Merge updated PTC inspect implementation into CUA output cleanup

.agents/notes/implemented/architecture/2026-11-09-explicit-ptc-output.i18n.yaml

@@ -9,11 +9,11 @@
   en: 51667474a3f9053d
   zh: 70c6af69e943f4f5
 /agent-note-explicit-text-and-image-output-in-ptc/decision:
-  en: c983f25d80de02e2
-  zh: 8f12dcb0b1ca1473
+  en: dc6138060f9dbe3f
+  zh: 22543a0d6774d61d
 /agent-note-explicit-text-and-image-output-in-ptc/alternatives-considered:
-  en: c36bfb244f527c7e
-  zh: 80c856d4809d2e25
+  en: d3fa023210117f56
+  zh: fbf9db7a7aa01a8e
 /agent-note-explicit-text-and-image-output-in-ptc/consequences:
   en: 636609d91f12a960
   zh: b26397c72f972827

.agents/notes/implemented/architecture/2026-11-09-explicit-ptc-output.md

@@ -10,7 +10,7 @@ PTC lets a program acquire and filter intermediate tool results before choosing
 
 ## Decision
 
-TypeScript `run_code` programs select text, objects and images through top-level `inspect(...values)`. Arguments are rendered independently: strings appear directly, and other lossless JSON values use two-space JSON formatting. Undefined arguments are omitted; invalid JSON values throw. Images can occur inside arrays or objects. One output sequence preserves the relationship between explanatory text and selected images.
+TypeScript `run_code` programs select text, objects and images through top-level `inspect(...values)`. Arguments are rendered independently: strings appear directly, and other values use Node's object formatting without depth, array-length or string-length preview limits. Inspection displays undefined properties, Error objects and circular references; requiring lossless JSON for display would interrupt subsequent work after tools have already executed. Images can occur inside arrays or objects. One output sequence preserves the relationship between explanatory text and selected images.
 
 The tool executor requests `discardCompletion: true`, so completion values are discarded before JSON validation and serialization. This is a consumer choice: direct runtime and workflow consumers retain completion values when their protocols require them. The [PTC foundation](../feature/2026-06-15-ptc.md) owns dispatch and settlement. The [typed-return decision](../feature/2026-07-20-ptc-typed-tool-returns.md) continues to own ordinary JSON binding values, failures and output accounting; this note owns TypeScript output selection and image projection.
 
@@ -30,7 +30,7 @@ Nested tool dispatches retain their complete rendered content in the Session log
 
 **Keep a CUA byte-returning facade and image sink.** This maintains a separate screenshot representation and a second image-emission path. Returning the runtime's Image lets file reads and screenshots share inspection and preserves the stored attachment reference through display.
 
-**Node inspection previews for selected values.** Item and string preview limits can omit file contents that a program explicitly selected. JSON formatting keeps those values complete within the shared output budget.
+**Bounded object previews.** Item and string preview limits can omit file contents that a program explicitly selected. Removing those limits keeps selected contents complete within the shared output budget.
 
 **Restrict console availability.** Missing bindings and incomplete replacements can break program code and imported libraries. The native console preserves Node methods while prompt guidance directs output selection through `inspect`.
 

.agents/notes/implemented/architecture/2026-11-09-explicit-ptc-output.zh.md

@@ -10,7 +10,7 @@ PTC 允许程序先获取和筛选中间工具结果，再选择模型看到的
 
 ## 决策
 
-TypeScript `run_code` 程序通过顶层 `inspect(...values)` 选择文本、对象和图片。各参数独立渲染：字符串直接展示，其他无损 JSON 值按两空格缩进的 JSON 格式展示。undefined 参数被省略；无效 JSON 值会抛出异常。图片也可以嵌入数组或对象。同一输出序列保留说明文字与所选图片的对应关系。
+TypeScript `run_code` 程序通过顶层 `inspect(...values)` 选择文本、对象和图片。各参数独立渲染：字符串直接展示，其他值使用 Node 的对象格式，不限制预览深度、数组项数或字符串长度。inspect 展示 undefined 属性、Error 对象和循环引用；要求展示值必须是无损 JSON，会在工具已经执行后中断后续工作。图片也可以嵌入数组或对象。同一输出序列保留说明文字与所选图片的对应关系。
 
 工具执行器请求 `discardCompletion: true`，在 JSON 校验和序列化前丢弃完成值。这是消费方的选择：直接运行时调用方和工作流消费方在其协议需要时保留完成值。[PTC 基础](../feature/2026-06-15-ptc.zh.md)负责分发与结算。[类型化返回值决策](../feature/2026-07-20-ptc-typed-tool-returns.zh.md)继续负责普通 JSON 绑定值、失败和输出计量；本记录负责 TypeScript 输出选择与图片投影。
 
@@ -30,7 +30,7 @@ Node 环境同时提供 `inspect` 和完整的原生 console。程序与导入
 
 **保留返回字节的 CUA 对象接口和图片输出回调。** 这会维护独立的截图表示和第二条图片输出路径。返回运行时 Image，使文件读取和截图共用展示方式，并在展示过程中保留已存储附件引用。
 
-**用 Node inspect 预览选中的值。** 条目和字符串的预览上限可能省略程序已明确选中的文件内容。JSON 格式在共享输出预算内完整保留这些值。
+**限制对象预览。** 条目和字符串的预览上限可能省略程序已明确选中的文件内容。取消这些限制，在共享输出预算内完整保留选中的内容。
 
 **限制 console 的可用性。** 缺失的绑定和不完整的替代实现可能使程序及导入的库出错。原生 console 保留 Node 方法，提示词指引程序通过 `inspect` 选择输出。
 

docs/upgrade-guide/v0.2.2-rc.1/ptc-inspect-output/guide.i18n.yaml

@@ -9,8 +9,8 @@
   en: 064ec0081b5c2946
   zh: 4b67a513ea949220
 /explicit-output-in-typescript-ptc-programs/change:
-  en: 7bce63e5291f5c31
-  zh: 6c13e850fde5d2ea
+  en: 37c78999dacb912a
+  zh: 1ff8f5605098ef59
 /explicit-output-in-typescript-ptc-programs/migration:
-  en: 2a2b58a6a09232be
-  zh: cdcd218579f9ed84
+  en: 5b97072ac3cce57f
+  zh: 136efac644c4bff4

docs/upgrade-guide/v0.2.2-rc.1/ptc-inspect-output/guide.md

@@ -9,7 +9,7 @@ English | [中文](guide.zh.md)
 
 ## Change
 
-TypeScript `run_code` programs display selected values through `inspect(...values)`. Function return values and final expressions previously appeared in the result; they are now omitted. Each inspected argument is rendered independently: strings appear directly, and other lossless JSON values use two-space JSON formatting. Undefined arguments are omitted. Values that cannot be represented as lossless JSON, such as binary views and cyclic objects, throw an exception.
+TypeScript `run_code` programs display selected values through `inspect(...values)`. Function return values and final expressions previously appeared in the result; they are now omitted. Each inspected argument is rendered independently: strings appear directly, and other values use Node's object formatting without depth, array-length or string-length preview limits. Undefined properties, Error objects and circular references remain visible.
 
 CUA observations previously produced output automatically. `getCapabilities()`, `listApps()`, `findApps(query)`, `getState()`, `app.getWindows()` and app or window `getAccessibilityTree()` now return their results to the program. Use `inspect` to display the values you need.
 
@@ -43,5 +43,4 @@ Other image-producing tools retain their declared program values and automatical
 3. Direct runtime consumers can set `discardCompletion: true` to select the same explicit-output behavior.
 4. Third-party `PtcRuntime` providers replace returned `logs` arrays with `output: PtcOutput[]`; direct runtime consumers read `result.output`. Text entries are `{ type: 'text', text }`, image entries are `{ type: 'image', attachment }`, and array order is display order.
 5. Upgrade preinstalled Node bootstraps selected by `launch.bootstrapPath` to protocol version 3 alongside the host. Readiness rejects incompatible revisions before program dispatch.
-6. Omit undefined properties from inspected containers. Convert non-JSON values to JSON values before inspection; for example, use `inspect(error.message)` to display an Error.
-7. Run an updated program and verify that inspected values appear once and its final expression adds no output.
+6. Run an updated program and verify that inspected values appear once and its final expression adds no output.

docs/upgrade-guide/v0.2.2-rc.1/ptc-inspect-output/guide.zh.md

@@ -9,7 +9,7 @@ description: "TypeScript PTC 程序通过 inspect 展示选定的值，并返回
 
 ## 变更
 
-TypeScript `run_code` 程序通过 `inspect(...values)` 展示选定的值。此前函数返回值和末尾表达式的值会进入结果，现在会被省略。各个 inspect 参数独立渲染：字符串直接展示，其他无损 JSON 值按两空格缩进的 JSON 格式展示。undefined 参数被省略。二进制视图、循环对象等无法表示为无损 JSON 的值会抛出异常。
+TypeScript `run_code` 程序通过 `inspect(...values)` 展示选定的值。此前函数返回值和末尾表达式的值会进入结果，现在会被省略。各个 inspect 参数独立渲染：字符串直接展示，其他值使用 Node 的对象格式，不限制预览深度、数组项数或字符串长度。undefined 属性、Error 对象和循环引用均会展示。
 
 此前，CUA 观察方法默认会输出结果。现在，`getCapabilities()`、`listApps()`、`findApps(query)`、`getState()`、`app.getWindows()` 以及应用或窗口的 `getAccessibilityTree()` 都将结果返回给程序。需要展示哪些值，就把它们传给 `inspect`。
 
@@ -43,5 +43,4 @@ TypeScript `run_code` 程序通过 `inspect(...values)` 展示选定的值。此
 3. 直接调用运行时的消费方可设置 `discardCompletion: true`，选择相同的显式输出行为。
 4. 第三方 `PtcRuntime` 提供方将返回的 `logs` 数组替换为 `output: PtcOutput[]`；直接消费运行时结果的调用方改为读取 `result.output`。文本条目为 `{ type: 'text', text }`，图片条目为 `{ type: 'image', attachment }`，数组顺序就是展示顺序。
 5. 随宿主一起将 `launch.bootstrapPath` 选用的预装 Node bootstrap 升级至协议版本 3。启动就绪检查会在派发程序前拒绝不兼容版本。
-6. 从待 inspect 的容器中省略 undefined 属性。将非 JSON 值转换为 JSON 值后再 inspect；例如，使用 `inspect(error.message)` 展示 Error。
-7. 运行更新后的程序，确认 inspect 的值只出现一次，且末尾表达式不额外输出。
+6. 运行更新后的程序，确认 inspect 的值只出现一次，且末尾表达式不额外输出。

packages/core/tool-executor-ptc/README.md

@@ -107,7 +107,7 @@ Inside the program:
 - Call tools as `await tools.name(args)` — quoted access for exotic names: `tools["my-tool"](args)`. Every call resolves to its declared value, which may be an Image. Tool arguments must be lossless JSON.
 - A FAILED tool call rejects with `ToolCallError`, whose `toolName` identifies the failed tool and whose `message` is human-readable — `try/catch` it to handle and continue.
 - Independent read-only calls MAY overlap under `Promise.all` (safe calls run concurrently; mutating calls run alone, in submission order). Sequence dependent work with `await`.
-- Show selected results with `inspect(...values)`. Each call accepts JSON values and Images, including Images nested in arrays and objects. Images expose a readable file path and its pixel dimensions. Extract just what you need.
+- Show selected values and images with `inspect(...values)`; every other intermediate result stays out of the conversation, so extract just what you need.
 
 Program-only SDK bindings:
 ```

packages/core/tool-executor-ptc/README.zh.md

@@ -107,7 +107,7 @@ Inside the program:
 - Call tools as `await tools.name(args)` — quoted access for exotic names: `tools["my-tool"](args)`. Every call resolves to its declared value, which may be an Image. Tool arguments must be lossless JSON.
 - A FAILED tool call rejects with `ToolCallError`, whose `toolName` identifies the failed tool and whose `message` is human-readable — `try/catch` it to handle and continue.
 - Independent read-only calls MAY overlap under `Promise.all` (safe calls run concurrently; mutating calls run alone, in submission order). Sequence dependent work with `await`.
-- Show selected results with `inspect(...values)`. Each call accepts JSON values and Images, including Images nested in arrays and objects. Images expose a readable file path and its pixel dimensions. Extract just what you need.
+- Show selected values and images with `inspect(...values)`; every other intermediate result stays out of the conversation, so extract just what you need.
 
 Program-only SDK bindings:
 ```

packages/core/tool-executor-ptc/src/ts-types.ts

@@ -264,7 +264,7 @@ function renderSdkProgramInstructions(): string {
 - Call tools as \`await tools.name(args)\` — quoted access for exotic names: \`tools["my-tool"](args)\`. Every call resolves to its declared value, which may be an Image. Tool arguments must be lossless JSON.
 - A FAILED tool call rejects with \`ToolCallError\`, whose \`toolName\` identifies the failed tool and whose \`message\` is human-readable — \`try/catch\` it to handle and continue.
 - Independent read-only calls MAY overlap under \`Promise.all\` (safe calls run concurrently; mutating calls run alone, in submission order). Sequence dependent work with \`await\`.
-- Show selected results with \`inspect(...values)\`. Each call accepts JSON values and Images, including Images nested in arrays and objects. Images expose a readable file path and its pixel dimensions. Extract just what you need.
+- Show selected values and images with \`inspect(...values)\`; every other intermediate result stays out of the conversation, so extract just what you need.
 
 Program-only SDK bindings:`
 }
@@ -312,6 +312,7 @@ export function renderToolsSdk(schemas: readonly ToolSdkSchema[], programFormat:
   for (const schema of sorted) {
     argsMembers.push(...docLines(schema.description, 1))
     argsMembers.push(`${pad(1)}${renderKey(schema.name)}: ${jsonSchemaToTs(schema.parameters, 1)};`)
+    // FIXME: Derive program return types from tool-declared adaptations, together with the binding adapter in ptc.ts.
     const output = schema.name === 'read_image' ? 'Image'
       : `${schema.name === 'cua' ? 'Image | ' : ''}${jsonSchemaToTs(schema.output, 1)}`
     outputMembers.push(`${pad(1)}${renderKey(schema.name)}: ${output};`)

packages/core/tool-executor-ptc/tests/ts-types.spec.ts

@@ -236,6 +236,7 @@ describe('renderToolsSdk', () => {
   it('uses explicit inspection in function-body mode', () => {
     const text = renderToolsSdk([bash])
     expect(text).toContain('inspect(await tools.bash(')
+    expect(text).toContain('- Show selected values and images with `inspect(...values)`; every other intermediate result stays out of the conversation, so extract just what you need.')
     expect(text).toContain('body of an async TypeScript function')
     expect(text).not.toContain('TypeScript REPL statements')
   })

packages/experimental/cua-macos/tests/large-image-ptc.spec.ts

@@ -164,7 +164,7 @@ it.skipIf(process.platform === 'win32')('delivers a valid PNG larger than 4 MiB
   expect(image.attachment.width).toBeGreaterThan(0)
   expect(image.attachment.width).toBeLessThan(WIDTH)
   expect(result.content.filter(block => block.type === 'text').map(block => block.text).join('\n'))
-    .toContain(`"width": ${image.attachment.width}`)
+    .toContain(`width: ${image.attachment.width}`)
   const stored = await context.attachments.readImage(image.attachment)
   const path = context.attachments.imageHostPath(image.attachment)
   if (path === undefined) throw new Error('The real local attachment has no disk path')

packages/ptc-runtime/ptc-runtime-node/README.i18n.yaml

@@ -21,8 +21,8 @@
   en: 8d42f921a1fa8e5c
   zh: 27b79ec4ec52b3fd
 /deepseek-ai-dsh-ptc-runtime-node/use-this-package/execution-and-results:
-  en: 721184490c958f6c
-  zh: b0580ceb07e59028
+  en: ae1a6eb24aac8ce9
+  zh: fd767cc70982e9b4
 /deepseek-ai-dsh-ptc-runtime-node/use-this-package/in-memory-modes:
   en: b731fd3de96dfbc8
   zh: 4891bcb431014678

packages/ptc-runtime/ptc-runtime-node/README.md

@@ -69,7 +69,7 @@ Profile editors can import `resolvePersistentExecution` from `@deepseek-ai/dsh-p
 
 In `per-call` mode, programs are async function bodies: top-level `await` and `return` work, and only erasable TypeScript is accepted. Node globals remain available, and a function body may declare a local `process`, including `const process = await import("node:process")`. A successful call returns its lossless-JSON value as `result.value` and captured text and images as ordered `result.output` entries. `result.sandbox` reports the selected mode, observed denial and the backend's full or partial enforcement independently of the program outcome.
 
-`inspect(...values)` accepts lossless JSON values and registered Images, including Images nested in arrays and objects. Strings render directly; other JSON values use two-space JSON formatting. Formatting follows the nesting limits of native `JSON.stringify`. Arrays and strings remain complete within the shared output budget. Undefined arguments are omitted. Omit undefined properties from containers and inspect `error.message` for Error objects. Invalid values, including binary views and cycles, throw an exception naming the argument; preceding inspected arguments remain visible. Text and selected images follow argument order.
+`inspect(...values)` renders strings directly and other values with Node's `util.inspect`, without depth, array-length or string-length preview limits. Undefined properties, Error objects and circular references remain visible. Registered Images can occur directly or inside arrays and objects; image discovery does not invoke getters or proxy traps. Text and selected images follow argument order within the shared output budget. If object formatting throws, preceding inspected arguments remain visible.
 
 All execution modes provide the complete native Node console, whose stdout and stderr writes enter the captured output. Model instructions say: "Use inspect(...values) to display output. Do not use console."
 

packages/ptc-runtime/ptc-runtime-node/README.zh.md

@@ -69,7 +69,7 @@ profile 编辑器可从 `@deepseek-ai/dsh-ptc-runtime-node/configuration` 导入
 
 `per-call` 模式中的程序是异步函数体：支持顶层 `await` 与 `return`，且只接受可擦除 TypeScript。Node 全局变量仍然可用，函数体可声明局部 `process`，例如 `const process = await import("node:process")`。成功调用以 `result.value` 返回无损 JSON 值，以 `result.output` 按顺序返回捕获的文本和图片。`result.sandbox` 独立于程序结果报告所选模式、观察到的拒绝，以及后端完整或部分的强制能力。
 
-`inspect(...values)` 接受无损 JSON 值和已注册的 Image，包括数组和对象中嵌套的 Image。字符串直接展示；其他 JSON 值按两空格缩进的 JSON 格式展示。格式化遵循原生 `JSON.stringify` 的嵌套深度限制。数组和字符串在共享输出预算内完整展示。undefined 参数被省略。容器中应省略 undefined 属性，Error 对象应改为 inspect `error.message`。二进制视图、循环引用等无效值会抛出指明参数位置的异常；此前已 inspect 的参数仍然可见。文本与选中的图片按参数顺序输出。
+`inspect(...values)` 直接渲染字符串，其他值使用 Node 的 `util.inspect` 格式，不限制预览深度、数组项数或字符串长度。undefined 属性、Error 对象和循环引用均会展示。注册的 Image 可以直接传入，也可以嵌入数组或对象；图片扫描不会调用 getter 或 Proxy trap。文本和所选图片在共享输出预算内遵循参数顺序。如果对象格式化抛错，此前 inspect 的参数仍然可见。
 
 所有执行模式都提供完整的 Node 原生 console，其 stdout 和 stderr 写入会进入捕获输出。模型指引要求：“使用 inspect(...values) 展示输出，不要使用 console。”
 

packages/ptc-runtime/ptc-runtime-node/src/bootstrap.ts

@@ -3,18 +3,20 @@
  * @module @deepseek-ai/dsh-ptc-runtime-node/src/bootstrap
  */
 
-import { types } from 'node:util'
+import { inspect, types } from 'node:util'
 import type { PtcJsonValue } from '@deepseek-ai/dsh-ptc-runtime'
 import type { HostMethods, ProgramResult, ProgramBootData } from './protocol.ts'
 import { jsonStringBytesUpTo, jsonValueBytesUpTo, truncateJsonStringBytes } from './output-json.ts'
 import { decodePtcJsonWire, encodePtcJsonWire, snapshotPtcJsonValue, type PtcImageWire } from './json-wire.ts'
 
 const CapturedError = Error
-const isNativeError = types.isNativeError
-const capturedJsonStringify = JSON.stringify
-const capturedObjectCreate = Object.create
-const capturedObjectDefineProperty = Object.defineProperty
-const capturedObjectKeys = Object.keys
+const { isNativeError, isArrayBufferView, isProxy } = types
+const {
+  create: capturedObjectCreate,
+  defineProperty: capturedObjectDefineProperty,
+  keys: capturedObjectKeys,
+  getOwnPropertyDescriptor: capturedObjectGetOwnPropertyDescriptor,
+} = Object
 
 /** Define one public binding-error field without consulting mutable globals or descriptor prototypes. */
 function defineBindingErrorField(error: Error, key: string, value: string): void {
@@ -94,12 +96,12 @@ export class LogBuffer {
 }
 
 /**
- * Render lossless JSON arguments independently and emit registered images in order.
+ * Render arguments without depth, array-length or string-length previews and emit registered images in order.
  * @param logs - the buffer every rendered line is pushed into.
  * @param images - Image identities owned by the execution process.
  * @param emitImage - Emits a registered image in the current evaluation.
  * @returns The program's inspect function.
- * @throws If a defined argument is not lossless JSON with optional registered Images, or native JSON formatting fails.
+ * @throws If object formatting fails; preceding arguments remain visible.
  */
 export function makeInspect(
   logs: Pick<LogBuffer, 'push'>, images?: Pick<WeakMap<object, string>, 'get'>, emitImage?: (id: string) => void,
@@ -111,43 +113,42 @@ export function makeInspect(
       logs.push(pending.join(' '))
       pending.length = 0
     }
-    for (const [index, arg] of args.entries()) {
-      const direct = arg !== null && typeof arg === 'object' ? images?.get(arg) : undefined
-      if (direct !== undefined && emitImage !== undefined) { flush(); emitImage(direct); continue }
-      if (arg === undefined) continue
-      let value: PtcJsonValue | undefined
-      try {
-        value = snapshotPtcJsonValue(arg, current => images?.get(current) !== undefined)
-      } catch (_error: unknown) {
-        // Accessors and proxy traps can prevent a lossless JSON snapshot.
-        value = undefined
-      }
-      if (value === undefined) {
+    for (const arg of args) {
+      const directImage = arg !== null && typeof arg === 'object' ? images?.get(arg) : undefined
+      if (directImage !== undefined && emitImage !== undefined) {
         flush()
-        throw new CapturedError(`inspect argument ${index + 1} must be lossless JSON or an Image; omit undefined properties and pass error.message for Errors`)
+        emitImage(directImage)
+        continue
       }
       try {
-        pending.push(typeof value === 'string' ? value : capturedJsonStringify(value, undefined, 2))
+        pending.push(typeof arg === 'string' ? arg : inspect(arg, { depth: null, maxArrayLength: null, maxStringLength: null }))
       } catch (error: unknown) {
         flush()
         throw error
       }
       if (images === undefined || emitImage === undefined) continue
       const nested: string[] = []
-      const visit: unknown[] = [value]
+      const seen = new Set<object>()
+      const visit: unknown[] = [arg]
       while (visit.length > 0) {
         const current = visit.pop()
         if (current === null || typeof current !== 'object') continue
         const id = images.get(current)
-        if (id !== undefined) { nested.push(id); continue }
+        if (id !== undefined) {
+          nested.push(id)
+          continue
+        }
+        // Binary contents contain no Images; proxy traps must not run during discovery.
+        if (seen.has(current) || isArrayBufferView(current) || isProxy(current)) continue
+        seen.add(current)
         for (const key of capturedObjectKeys(current).reverse()) {
-          visit.push((current as Record<string, unknown>)[key])
+          const property = capturedObjectGetOwnPropertyDescriptor(current, key)
+          if (property !== undefined && 'value' in property) visit.push(property.value)
         }
       }
-      if (nested.length > 0) {
-        flush()
-        for (const id of nested) emitImage(id)
-      }
+      if (nested.length === 0) continue
+      flush()
+      for (const id of nested) emitImage(id)
     }
     flush()
     if (args.length === 0) logs.push('')

packages/ptc-runtime/ptc-runtime-node/tests/bootstrap.spec.ts

@@ -36,47 +36,37 @@ describe('makeInspect', () => {
     const seen: string[] = []
     const inspect = makeInspect(new LogBuffer(1_000, text => seen.push(text)))
     inspect('plain', { a: 1 })
-    expect(seen).toEqual(['plain {\n  "a": 1\n}'])
+    expect(seen).toEqual(['plain { a: 1 }'])
   })
 
   it.each([
-    ['Buffer', Buffer.from([1, 2, 3])],
-    ['TypedArray', new Uint8Array([1, 2, 3])],
-    ['DataView', new DataView(new ArrayBuffer(3))],
-    ['ArrayBuffer', new ArrayBuffer(3)],
-    ['function', () => 1],
-    ['Error', new Error('inspection failure')],
-    ['bigint', 1n],
-    ['non-finite number', Infinity],
-    ['negative zero', -0],
-    ['undefined property', { value: undefined }],
-  ])('rejects a %s argument after preserving preceding text', (_name, value) => {
+    ['Buffer', Buffer.from([1, 2, 3]), '<Buffer 01 02 03>'],
+    ['TypedArray', new Uint8Array([1, 2, 3]), 'Uint8Array(3) [ 1, 2, 3 ]'],
+    ['function', () => 1, '[Function'],
+    ['Error', new Error('inspection failure'), 'Error: inspection failure'],
+    ['bigint', 1n, '1n'],
+    ['non-finite number', Infinity, 'Infinity'],
+    ['negative zero', -0, '-0'],
+    ['undefined property', { value: undefined }, '{ value: undefined }'],
+  ])('displays a %s argument and continues execution', (_name, value, expected) => {
     const text: string[] = []
     const inspect = makeInspect({ push: line => text.push(line) })
-    expect(() => { inspect('before', value, 'after') }).toThrow('must be lossless JSON or an Image')
-    expect(text).toEqual(['before'])
+    inspect('before', value, 'after')
+    expect(text).toHaveLength(1)
+    expect(text[0]).toContain(expected)
+    expect(text[0]).toMatch(/^before .* after$/s)
   })
 
-  it('reports revoked proxies and throwing accessors as invalid inspected values', () => {
+  it('displays revoked proxies and accessors without invoking them', () => {
     const revoked = Proxy.revocable({}, {})
     revoked.revoke()
-    const throwing = { get value() { throw new Error('getter') } }
-    for (const value of [revoked.proxy, throwing]) {
-      const inspect = makeInspect({ push() {} })
-      expect(() => { inspect(value) }).toThrow('must be lossless JSON or an Image')
-    }
-  })
-
-  it.each([{ code: undefined }, new Error('tool failed')])('identifies the invalid argument and explains JSON conversion', (value) => {
     const text: string[] = []
-    const inspect = makeInspect(new LogBuffer(1_000, line => text.push(line)))
-    expect(() => { inspect('before', value) }).toThrow(
-      'inspect argument 2 must be lossless JSON or an Image; omit undefined properties and pass error.message for Errors',
-    )
-    expect(text).toEqual(['before'])
+    const inspect = makeInspect({ push: line => text.push(line) }, new WeakMap(), () => {})
+    inspect(revoked.proxy, { get value() { throw new Error('getter') } })
+    expect(text).toEqual(['<Revoked Proxy> { value: [Getter] }'])
   })
 
-  it('renders complete arrays, long strings and nested JSON values', () => {
+  it('renders complete arrays, long strings and deeply nested objects', () => {
     const value = {
       lines: Array.from({ length: 330 }, (_, index) => ({ number: index + 1, text: `line ${index + 1}` })),
       text: 'x'.repeat(12_000),
@@ -85,15 +75,14 @@ describe('makeInspect', () => {
     const text: string[] = []
     makeInspect(new LogBuffer(100_000, line => text.push(line)))(value)
     expect(text).toHaveLength(1)
-    expect(JSON.parse(text[0]!)).toEqual(value)
+    expect(runInNewContext(`(${text[0]!})`)).toEqual(value)
   })
 
-  it('preserves preceding arguments when native JSON formatting fails', () => {
-    let value: unknown = 0
-    for (let depth = 0; depth < 10_000; depth++) value = { child: value }
+  it('preserves preceding arguments when a custom formatter throws', () => {
+    const value = { [Symbol.for('nodejs.util.inspect.custom')]() { throw new Error('formatter failed') } }
     const text: string[] = []
     const inspect = makeInspect(new LogBuffer(1_000, line => text.push(line)))
-    expect(() => { inspect('before', value, 'after') }).toThrow(RangeError)
+    expect(() => { inspect('before', value, 'after') }).toThrow('formatter failed')
     expect(text).toEqual(['before'])
   })
 
@@ -107,38 +96,49 @@ describe('makeInspect', () => {
     const inspect = makeInspect({ push: line => text.push(line) }, images, id => selected.push(id))
     inspect('before', value, 'after', one)
     expect(selected).toEqual(['one', 'two', 'one'])
-    expect(JSON.parse(text[0]!.slice('before '.length))).toEqual(value)
+    expect(runInNewContext(`(${text[0]!.slice('before '.length)})`)).toEqual(value)
     expect(text[1]).toBe('after')
     inspect({ path: '/fake.png', width: 1, height: 1 }, null)
     inspect()
     expect(selected).toEqual(['one', 'two', 'one'])
-    expect(text.slice(2)).toEqual(['{\n  "path": "/fake.png",\n  "width": 1,\n  "height": 1\n} null', ''])
+    expect(text.slice(2)).toEqual(["{ path: '/fake.png', width: 1, height: 1 } null", ''])
     makeInspect({ push: line => text.push(line) }, images)(one)
-    expect(JSON.parse(text.at(-1)!)).toEqual(one)
+    expect(runInNewContext(`(${text.at(-1)!})`)).toEqual(one)
   })
 
-  it('uses one JSON snapshot for text and image selection', () => {
+  it('does not select images behind getters', () => {
     const image = Object.freeze({ path: '/image.png', width: 8, height: 4 })
     const images = new WeakMap<object, string>([[image, 'image']])
     const read = vi.fn(() => image)
     const text: string[] = []
     const selected: string[] = []
     makeInspect({ push: line => text.push(line) }, images, id => selected.push(id))({ get image() { return read() } })
-    expect(read).toHaveBeenCalledTimes(1)
-    expect(JSON.parse(text[0]!)).toEqual({ image })
+    expect(read).not.toHaveBeenCalled()
+    expect(text).toEqual(['{ image: [Getter] }'])
+    expect(selected).toEqual([])
+  })
+
+  it('finds images in cyclic containers alongside binary views', () => {
+    const image = Object.freeze({ path: '/image.png', width: 8, height: 4 })
+    const images = new WeakMap<object, string>([[image, 'image']])
+    const value: Record<string, unknown> = { binary: new Uint8Array([1, 2, 3]), image }
+    value.self = value
+    const selected: string[] = []
+    makeInspect({ push() {} }, images, id => selected.push(id))(value)
     expect(selected).toEqual(['image'])
   })
 
-  it('omits undefined arguments and rejects cycles', () => {
+  it('displays undefined arguments and cycles', () => {
     const text: string[] = []
     const inspect = makeInspect({ push: line => text.push(line) })
     inspect(undefined)
-    expect(text).toEqual([])
+    expect(text).toEqual(['undefined'])
     inspect(null)
-    expect(text).toEqual(['null'])
+    expect(text).toEqual(['undefined', 'null'])
     const cyclic: Record<string, unknown> = {}
     cyclic.self = cyclic
-    expect(() => { inspect(cyclic) }).toThrow('must be lossless JSON or an Image')
+    inspect(cyclic)
+    expect(text.at(-1)).toBe('<ref *1> { self: [Circular *1] }')
   })
 })
 

packages/ptc-runtime/ptc-runtime-node/tests/output.spec.ts

@@ -1,4 +1,5 @@
 import { AttachmentId } from '@deepseek-ai/dsh-attachment'
+import { runInNewContext } from 'node:vm'
 import { PtcImage } from '@deepseek-ai/dsh-ptc-runtime'
 import { Context } from '@deepseek-ai/cordis'
 import { describe, expect, it } from 'vitest'
@@ -26,23 +27,33 @@ describe.each(modes)('program output in $executionMode with retention $scriptPer
     const { run } = await setup()
     const result = await run('inspect("%s %d %j %%", "text", 7, { n: 1 }, [2, 3]);')
     expect(result.error).toBeUndefined()
-    expect(result.output.filter(block => block.type === 'text').map(block => block.text)).toEqual(['%s %d %j %% text 7 {\n  "n": 1\n} [\n  2,\n  3\n]'])
+    expect(result.output.filter(block => block.type === 'text').map(block => block.text)).toEqual(['%s %d %j %% text 7 { n: 1 } [ 2, 3 ]'])
     expect(result.value).toBeUndefined()
   })
 
-  it.each(['Buffer.alloc(16 * 1024 * 1024)', 'new Uint8Array(16 * 1024 * 1024)'])('rejects non-JSON binary output: %s', async (value) => {
-    const { run } = await setup()
-    const result = await run(`
-      const { Buffer } = await import('node:buffer');
-      inspect('before', ${value});
-      inspect('after');
-    `)
-    expect(result.error?.kind).toBe('exception')
-    expect(result.error?.message).toContain('must be lossless JSON or an Image')
-    expect(result.output).toEqual([{ type: 'text', text: 'before' }])
+  it('continues to the next binding after inspecting a missing result field and an Error', async () => {
+    const { runtime } = await setup()
+    const calls: string[] = []
+    const result = await runtime.run(runtime.resolve({
+      program: `
+        const result = await tools.edit({});
+        inspect({ path: result.path, operation: result.operation });
+        inspect(new Error('inspection detail'));
+        await tools.test({});
+      `,
+      bindings: [{ global: 'tools', functions: {
+        edit: () => { calls.push('edit'); return Promise.resolve({ path: '/file.ts', before: 'before', after: 'after' }) },
+        test: () => { calls.push('test'); return Promise.resolve(null) },
+      } }],
+      discardCompletion: true,
+    }))
+    expect(result.error).toBeUndefined()
+    expect(calls).toEqual(['edit', 'test'])
+    expect(result.output[0]).toEqual({ type: 'text', text: "{ path: '/file.ts', operation: undefined }" })
+    expect(result.output[1]).toEqual({ type: 'text', text: expect.stringContaining('Error: inspection detail') as string })
   })
 
-  it('reports invalid JSON inspection and accepts a subsequent program', async () => {
+  it('displays a revoked proxy and continues execution', async () => {
     const { run } = await setup()
     const result = await run(`
       const revoked = Proxy.revocable({}, {});
@@ -50,15 +61,14 @@ describe.each(modes)('program output in $executionMode with retention $scriptPer
       inspect('before', revoked.proxy);
       inspect('after');
     `)
-    expect(result.error?.kind).toBe('exception')
-    expect(result.error?.message).toContain('must be lossless JSON or an Image')
-    expect(result.output).toEqual([{ type: 'text', text: 'before' }])
+    expect(result.error).toBeUndefined()
+    expect(result.output).toEqual([{ type: 'text', text: 'before <Revoked Proxy>' }, { type: 'text', text: 'after' }])
     const recovered = await run('inspect("recovered")')
     expect(recovered.error).toBeUndefined()
     expect(recovered.output).toEqual([{ type: 'text', text: 'recovered' }])
   })
 
-  it('displays complete JSON arrays and long strings', async () => {
+  it('displays complete arrays, long strings and deeply nested objects', async () => {
     const { run } = await setup()
     const result = await run(`inspect({
       lines: Array.from({ length: 330 }, (_, index) => ({ number: index + 1, text: "line " + (index + 1) })),
@@ -67,8 +77,8 @@ describe.each(modes)('program output in $executionMode with retention $scriptPer
     })`)
     expect(result.error).toBeUndefined()
     const text = result.output[0]
-    if (text?.type !== 'text') throw new Error('Expected JSON output')
-    expect(JSON.parse(text.text)).toEqual({
+    if (text?.type !== 'text') throw new Error('Expected inspected output')
+    expect(runInNewContext(`(${text.text})`)).toEqual({
       lines: Array.from({ length: 330 }, (_, index) => ({ number: index + 1, text: `line ${index + 1}` })),
       text: 'x'.repeat(12_000),
       nested: { a: { b: { c: { d: { e: { answer: 42 } } } } } },
@@ -76,11 +86,11 @@ describe.each(modes)('program output in $executionMode with retention $scriptPer
     expect(result.output).toHaveLength(1)
   })
 
-  it('preserves preceding text when native JSON formatting fails', async () => {
+  it('preserves preceding text when a custom formatter throws', async () => {
     const { run } = await setup()
-    const result = await run('let value = 0; for (let i = 0; i < 10000; i++) value = [value]; inspect("before", value)')
+    const result = await run('inspect("before", { [Symbol.for("nodejs.util.inspect.custom")]() { throw new Error("formatter failed") } })')
     expect(result.error?.kind).toBe('exception')
-    expect(result.error?.message).toContain('Maximum call stack size exceeded')
+    expect(result.error?.message).toContain('formatter failed')
     expect(result.output).toEqual([{ type: 'text', text: 'before' }])
   })
 
@@ -107,17 +117,17 @@ describe.each(modes)('program output in $executionMode with retention $scriptPer
     expect(result.error).toBeUndefined()
     expect(result.output).toEqual([
       { type: 'text', text: 'true sha256:before /images/before.png 8 4 true' },
-      { type: 'text', text: 'before {\n  "windowId": 23\n}' },
+      { type: 'text', text: 'before { windowId: 23 }' },
       { type: 'image', attachment: before.attachment },
       { type: 'text', text: expect.stringContaining('after [') as string },
       { type: 'image', attachment: after.attachment },
       { type: 'image', attachment: before.attachment },
-      { type: 'text', text: expect.stringContaining('"forged":') as string },
+      { type: 'text', text: expect.stringContaining('forged:') as string },
     ])
     const preview = result.output[3]
     if (preview?.type !== 'text') throw new Error('Expected the nested image preview')
-    expect(preview.text).toContain('"attachmentId": "sha256:before"')
-    expect(preview.text).toContain('"attachmentId": "sha256:after"')
+    expect(preview.text).toContain("attachmentId: 'sha256:before'")
+    expect(preview.text).toContain("attachmentId: 'sha256:after'")
     expect(result.output.find(block => block.type === 'image')?.attachment).toBe(before.attachment)
   })
 
@@ -174,15 +184,15 @@ describe.each(modes)('program output in $executionMode with retention $scriptPer
     expect(selected.output).toEqual([])
   })
 
-  it('retains an inspected JSON prefix within the outer output budget', async () => {
+  it('retains an inspected prefix within the outer output budget', async () => {
     const { runtime } = await mountRuntime(new Context(), { ...mode, maxOutputBytes: 128 })
     const result = await runtime.run(runtime.resolve({
       program: 'inspect({ value: "x".repeat(512) })', bindings: [], discardCompletion: true,
     }))
     expect(result.error?.kind).toBe('output-limit')
     const prefix = result.output[0]
-    if (prefix?.type !== 'text') throw new Error('Expected a JSON text prefix')
-    expect(prefix.text).toMatch(/^\{\n  "value": "x+/)
+    if (prefix?.type !== 'text') throw new Error('Expected an inspected text prefix')
+    expect(prefix.text).toMatch(/^\{\s+value: 'x+/)
     expect(Buffer.byteLength(prefix.text)).toBeLessThan(128)
   })
 

snapshots/sdk/desktop-ptc-context/system-prompt.expected.md

@@ -47,7 +47,7 @@ Inside the program:
 - Call tools as `await tools.name(args)` — quoted access for exotic names: `tools["my-tool"](args)`. Every call resolves to its declared value, which may be an Image. Tool arguments must be lossless JSON.
 - A FAILED tool call rejects with `ToolCallError`, whose `toolName` identifies the failed tool and whose `message` is human-readable — `try/catch` it to handle and continue.
 - Independent read-only calls MAY overlap under `Promise.all` (safe calls run concurrently; mutating calls run alone, in submission order). Sequence dependent work with `await`.
-- Show selected results with `inspect(...values)`. Each call accepts JSON values and Images, including Images nested in arrays and objects. Images expose a readable file path and its pixel dimensions. Extract just what you need.
+- Show selected values and images with `inspect(...values)`; every other intermediate result stays out of the conversation, so extract just what you need.
 
 Program-only SDK bindings:
 

snapshots/sdk/tool-error-details/system-prompt.expected.md

@@ -83,7 +83,7 @@ Inside the program:
 - Call tools as `await tools.name(args)` — quoted access for exotic names: `tools["my-tool"](args)`. Every call resolves to its declared value, which may be an Image. Tool arguments must be lossless JSON.
 - A FAILED tool call rejects with `ToolCallError`, whose `toolName` identifies the failed tool and whose `message` is human-readable — `try/catch` it to handle and continue.
 - Independent read-only calls MAY overlap under `Promise.all` (safe calls run concurrently; mutating calls run alone, in submission order). Sequence dependent work with `await`.
-- Show selected results with `inspect(...values)`. Each call accepts JSON values and Images, including Images nested in arrays and objects. Images expose a readable file path and its pixel dimensions. Extract just what you need.
+- Show selected values and images with `inspect(...values)`; every other intermediate result stays out of the conversation, so extract just what you need.
 
 Program-only SDK bindings:
 

snapshots/session/advanced-toolchain/system-prompt.expected.md

@@ -83,7 +83,7 @@ Inside the program:
 - Call tools as `await tools.name(args)` — quoted access for exotic names: `tools["my-tool"](args)`. Every call resolves to its declared value, which may be an Image. Tool arguments must be lossless JSON.
 - A FAILED tool call rejects with `ToolCallError`, whose `toolName` identifies the failed tool and whose `message` is human-readable — `try/catch` it to handle and continue.
 - Independent read-only calls MAY overlap under `Promise.all` (safe calls run concurrently; mutating calls run alone, in submission order). Sequence dependent work with `await`.
-- Show selected results with `inspect(...values)`. Each call accepts JSON values and Images, including Images nested in arrays and objects. Images expose a readable file path and its pixel dimensions. Extract just what you need.
+- Show selected values and images with `inspect(...values)`; every other intermediate result stays out of the conversation, so extract just what you need.
 
 Program-only SDK bindings:
 

snapshots/session/both-mode-turn/system-prompt.expected.md

@@ -46,7 +46,7 @@ Inside the program:
 - Call tools as `await tools.name(args)` — quoted access for exotic names: `tools["my-tool"](args)`. Every call resolves to its declared value, which may be an Image. Tool arguments must be lossless JSON.
 - A FAILED tool call rejects with `ToolCallError`, whose `toolName` identifies the failed tool and whose `message` is human-readable — `try/catch` it to handle and continue.
 - Independent read-only calls MAY overlap under `Promise.all` (safe calls run concurrently; mutating calls run alone, in submission order). Sequence dependent work with `await`.
-- Show selected results with `inspect(...values)`. Each call accepts JSON values and Images, including Images nested in arrays and objects. Images expose a readable file path and its pixel dimensions. Extract just what you need.
+- Show selected values and images with `inspect(...values)`; every other intermediate result stays out of the conversation, so extract just what you need.
 
 Program-only SDK bindings:
 

snapshots/session/cua-ptc/session.v4.jsonl

@@ -22,7 +22,7 @@
 {"type":"tool/ptc-dispatch","data":{"rootCallId":"cua-ptc-native-result","parentCallId":"cua-ptc-native-result","subCallId":"cua-ptc-native-result:ptc:3","name":"cua","arguments":{"method":"getState","params":{"appHandle":"fixture-app","windowId":7,"maxElements":16}},"isError":false,"content":[{"type":"text","text":"{\"appHandle\":\"fixture-app\",\"windowId\":7,\"focusedWindowId\":7,\"snapshotId\":\"fixture-snapshot-2\",\"tree\":\"AX scope=window offsets=UTF-16\\n- [0] AXButton \\\"Increment\\\"\\n- [1] AXTextField valueWritable=true selectionWritable=true value=\\\"\\\" placeholder=\\\"Fixture text\\\"\\n- AXStaticText \\\"Count: 0\\\"\\n- [2] AXMenuBarItem \\\"File\\\"\\nMenu contents omitted; request includeMenus=true to expand.\",\"windows\":[{\"windowId\":7,\"appName\":\"Fixture\",\"title\":\"CUA fixture\",\"bounds\":{\"x\":40,\"y\":60,\"width\":320,\"height\":200},\"isOnScreen\":true,\"onCurrentSpace\":true},{\"windowId\":8,\"appName\":\"Fixture\",\"title\":\"AXless canvas\",\"bounds\":{\"x\":40,\"y\":60,\"width\":320,\"height\":200},\"isOnScreen\":true,\"onCurrentSpace\":true},{\"windowId\":9,\"appName\":\"Fixture\",\"title\":\"Transformed window\",\"bounds\":{\"x\":16,\"y\":512,\"width\":131,\"height\":145},\"isOnScreen\":true,\"onCurrentSpace\":true}],\"truncated\":false}"}]}}
 {"type":"tool/ptc-dispatch-start","data":{"rootCallId":"cua-ptc-native-result","parentCallId":"cua-ptc-native-result","subCallId":"cua-ptc-native-result:ptc:4","name":"cua","arguments":{"method":"screenshot","params":{"appHandle":"fixture-app","windowId":7}}}}
 {"type":"tool/ptc-dispatch","data":{"rootCallId":"cua-ptc-native-result","parentCallId":"cua-ptc-native-result","subCallId":"cua-ptc-native-result:ptc:4","name":"cua","arguments":{"method":"screenshot","params":{"appHandle":"fixture-app","windowId":7}},"isError":false,"content":[{"type":"text","text":"{\"pixelWidth\":1,\"pixelHeight\":1,\"windowId\":7,\"bounds\":{\"x\":40,\"y\":60,\"width\":320,\"height\":200},\"coordinateSpace\":\"screenshot-pixels\"}"},{"type":"image","attachment":{"attachmentId":"sha256:999f1d1527ee7e79266f16add5430fff76b1225d742464a5b1ff1f02971bb8ee","mediaType":"image/png","width":1,"height":1,"bytes":90}}]}}
-{"type":"tool/result","data":{"turn":1,"step":1,"message":{"role":"tool","source":{"kind":"tool","callId":"cua-ptc-native-result"},"toolCallId":"cua-ptc-native-result","content":[{"type":"text","text":"{\n  \"appHandle\": \"fixture-app\",\n  \"initiallyTruncated\": true,\n  \"recoveredTextField\": true,\n  \"hasIncrement\": true,\n  \"pngBytes\": 90,\n  \"signature\": \"89504e470d0a1a0a\",\n  \"imageWidth\": 1,\n  \"imageHeight\": 1\n}"},{"type":"image","attachment":{"attachmentId":"sha256:999f1d1527ee7e79266f16add5430fff76b1225d742464a5b1ff1f02971bb8ee","mediaType":"image/png","width":1,"height":1,"bytes":90}}],"isError":false,"id":"{{message:5}}"}},"sourceEventSeqs":[14],"surfaceOp":"append"}
+{"type":"tool/result","data":{"turn":1,"step":1,"message":{"role":"tool","source":{"kind":"tool","callId":"cua-ptc-native-result"},"toolCallId":"cua-ptc-native-result","content":[{"type":"text","text":"{\n  appHandle: 'fixture-app',\n  initiallyTruncated: true,\n  recoveredTextField: true,\n  hasIncrement: true,\n  pngBytes: 90,\n  signature: '89504e470d0a1a0a',\n  imageWidth: 1,\n  imageHeight: 1\n}"},{"type":"image","attachment":{"attachmentId":"sha256:999f1d1527ee7e79266f16add5430fff76b1225d742464a5b1ff1f02971bb8ee","mediaType":"image/png","width":1,"height":1,"bytes":90}}],"isError":false,"id":"{{message:5}}"}},"sourceEventSeqs":[14],"surfaceOp":"append"}
 {"type":"step/end","data":{"turn":1,"step":1}}
 {"type":"step/start","data":{"turn":1,"step":2}}
 {"type":"assistant/message","data":{"turn":1,"step":2,"message":{"role":"assistant","content":[{"type":"tool-call","id":"cua-ptc-ax-failure","name":"run_code","arguments":"{\"code\":\"const { readFile, writeFile, unlink } = await import('node:fs/promises');\\nconst { resolve } = await import('node:path');\\nconst { pathToFileURL } = await import('node:url');\\nconst { connectCua } = await import(pathToFileURL(resolve('cua-bridge.mjs')).href);\\nawait writeFile('fixture-ax-failure', '');\\ntry {\\n  const cua = connectCua(() => tools.cua);\\n  const app = await cua.getApp('test.fixture');\\n  const windows = await app.getWindows();\\n  const window = await app.getWindow(windows[0].windowId);\\n  const failures = [];\\n  for (const surface of [app, window]) {\\n    try { await surface.getAccessibilityTree(); throw new Error('Expected AX failure'); }\\n    catch (error) { if (!error.message.includes('Fixture accessibility read failed')) throw error; failures.push(error.message); }\\n  }\\n  let coordinateRefused = false;\\n  try { await window.click([100, 100], { relative: true }); }\\n  catch (error) { if (!error.message.includes('Window has no observation')) throw error; coordinateRefused = true; }\\n  if (!coordinateRefused) throw new Error('Coordinate input accepted without observation');\\n  const shot = await window.getScreenshot();\\n  const coordinateDispatched = (await window.click([100, 100], { relative: true })).dispatched;\\n  let elementRefused = false;\\n  try { await window.invoke('0'); }\\n  catch (error) { if (!error.message.includes('Read getAccessibilityTree')) throw error; elementRefused = true; }\\n  if (!elementRefused) throw new Error('Element input accepted without an AX snapshot');\\n  inspect({ axFailures: failures.length, screenshotWindow: window.windowId, pngBytes: (await readFile(shot.path)).byteLength, elementRefused, coordinateRefused, coordinateDispatched }, shot);\\n} finally { await unlink('fixture-ax-failure'); }\",\"description\":\"Capture through the object API despite application and window AX failures\"}"}],"source":{"kind":"model","provider":"deepseek-official","model":"deepseek-v4-flash-vision-exp"},"id":"{{message:6}}"},"usage":{"inputTokens":3,"outputTokens":3},"stream":[{"type":"chunk","time":1794226987900,"chunk":{"type":"block-start","index":0,"blockType":"tool-call"}},{"type":"chunk","time":1794226987900,"chunk":{"type":"block-end","index":0,"block":{"type":"tool-call","id":"cua-ptc-ax-failure","name":"run_code","arguments":"{\"code\":\"const { readFile, writeFile, unlink } = await import('node:fs/promises');\\nconst { resolve } = await import('node:path');\\nconst { pathToFileURL } = await import('node:url');\\nconst { connectCua } = await import(pathToFileURL(resolve('cua-bridge.mjs')).href);\\nawait writeFile('fixture-ax-failure', '');\\ntry {\\n  const cua = connectCua(() => tools.cua);\\n  const app = await cua.getApp('test.fixture');\\n  const windows = await app.getWindows();\\n  const window = await app.getWindow(windows[0].windowId);\\n  const failures = [];\\n  for (const surface of [app, window]) {\\n    try { await surface.getAccessibilityTree(); throw new Error('Expected AX failure'); }\\n    catch (error) { if (!error.message.includes('Fixture accessibility read failed')) throw error; failures.push(error.message); }\\n  }\\n  let coordinateRefused = false;\\n  try { await window.click([100, 100], { relative: true }); }\\n  catch (error) { if (!error.message.includes('Window has no observation')) throw error; coordinateRefused = true; }\\n  if (!coordinateRefused) throw new Error('Coordinate input accepted without observation');\\n  const shot = await window.getScreenshot();\\n  const coordinateDispatched = (await window.click([100, 100], { relative: true })).dispatched;\\n  let elementRefused = false;\\n  try { await window.invoke('0'); }\\n  catch (error) { if (!error.message.includes('Read getAccessibilityTree')) throw error; elementRefused = true; }\\n  if (!elementRefused) throw new Error('Element input accepted without an AX snapshot');\\n  inspect({ axFailures: failures.length, screenshotWindow: window.windowId, pngBytes: (await readFile(shot.path)).byteLength, elementRefused, coordinateRefused, coordinateDispatched }, shot);\\n} finally { await unlink('fixture-ax-failure'); }\",\"description\":\"Capture through the object API despite application and window AX failures\"}"}}},{"type":"chunk","time":1794226987901,"chunk":{"type":"usage","usage":{"inputTokens":3,"outputTokens":3}}},{"type":"chunk","time":1794226987901,"chunk":{"type":"finish","reason":{"kind":"tool-calls"}}}]},"surfaceOp":"append"}
@@ -43,7 +43,7 @@
 {"type":"tool/ptc-dispatch","data":{"rootCallId":"cua-ptc-ax-failure","parentCallId":"cua-ptc-ax-failure","subCallId":"cua-ptc-ax-failure:ptc:7","name":"cua","arguments":{"method":"screenshot","params":{"appHandle":"fixture-app","windowId":7}},"isError":false,"content":[{"type":"text","text":"{\"pixelWidth\":1,\"pixelHeight\":1,\"windowId\":7,\"bounds\":{\"x\":40,\"y\":60,\"width\":320,\"height\":200},\"coordinateSpace\":\"screenshot-pixels\"}"},{"type":"image","attachment":{"attachmentId":"sha256:999f1d1527ee7e79266f16add5430fff76b1225d742464a5b1ff1f02971bb8ee","mediaType":"image/png","width":1,"height":1,"bytes":90}}]}}
 {"type":"tool/ptc-dispatch-start","data":{"rootCallId":"cua-ptc-ax-failure","parentCallId":"cua-ptc-ax-failure","subCallId":"cua-ptc-ax-failure:ptc:8","name":"cua","arguments":{"method":"click","params":{"appHandle":"fixture-app","windowId":7,"x":100,"y":100,"relative":true,"mouseButton":"left","clickCount":1}}}}
 {"type":"tool/ptc-dispatch","data":{"rootCallId":"cua-ptc-ax-failure","parentCallId":"cua-ptc-ax-failure","subCallId":"cua-ptc-ax-failure:ptc:8","name":"cua","arguments":{"method":"click","params":{"appHandle":"fixture-app","windowId":7,"x":100,"y":100,"relative":true,"mouseButton":"left","clickCount":1}},"isError":false,"content":[{"type":"text","text":"{\"dispatched\":true}"}]}}
-{"type":"tool/result","data":{"turn":1,"step":2,"message":{"role":"tool","source":{"kind":"tool","callId":"cua-ptc-ax-failure"},"toolCallId":"cua-ptc-ax-failure","content":[{"type":"text","text":"{\n  \"axFailures\": 2,\n  \"screenshotWindow\": 7,\n  \"pngBytes\": 90,\n  \"elementRefused\": true,\n  \"coordinateRefused\": true,\n  \"coordinateDispatched\": true\n}"},{"type":"image","attachment":{"attachmentId":"sha256:999f1d1527ee7e79266f16add5430fff76b1225d742464a5b1ff1f02971bb8ee","mediaType":"image/png","width":1,"height":1,"bytes":90}}],"isError":false,"id":"{{message:7}}"}},"sourceEventSeqs":[27],"surfaceOp":"append"}
+{"type":"tool/result","data":{"turn":1,"step":2,"message":{"role":"tool","source":{"kind":"tool","callId":"cua-ptc-ax-failure"},"toolCallId":"cua-ptc-ax-failure","content":[{"type":"text","text":"{\n  axFailures: 2,\n  screenshotWindow: 7,\n  pngBytes: 90,\n  elementRefused: true,\n  coordinateRefused: true,\n  coordinateDispatched: true\n}"},{"type":"image","attachment":{"attachmentId":"sha256:999f1d1527ee7e79266f16add5430fff76b1225d742464a5b1ff1f02971bb8ee","mediaType":"image/png","width":1,"height":1,"bytes":90}}],"isError":false,"id":"{{message:7}}"}},"sourceEventSeqs":[27],"surfaceOp":"append"}
 {"type":"step/end","data":{"turn":1,"step":2}}
 {"type":"step/start","data":{"turn":1,"step":3}}
 {"type":"assistant/message","data":{"turn":1,"step":3,"message":{"role":"assistant","content":[{"type":"tool-call","id":"cua-ptc-writability","name":"run_code","arguments":"{\"code\":\"const { writeFile, unlink } = await import('node:fs/promises');\\nconst { resolve } = await import('node:path');\\nconst { pathToFileURL } = await import('node:url');\\nconst { connectCua } = await import(pathToFileURL(resolve('cua-bridge.mjs')).href);\\nconst cua = connectCua(() => tools.cua);\\nconst app = await cua.getApp('test.fixture');\\nconst results = [];\\nfor (const attribute of ['AXValue', 'AXSelectedTextRange']) {\\n  await writeFile('fixture-ax-writability-failure', attribute);\\n  try {\\n    const tree = await app.getAccessibilityTree();\\n    const field = attribute === 'AXValue' ? 'valueWritable' : 'selectionWritable';\\n    if (!tree.includes(field + '=unknown') || tree.includes(field + '=true')) throw new Error('Unknown writability was lost');\\n    if (!tree.includes('Increment') || !tree.includes('Count:')) throw new Error('Other nodes were lost');\\n    let writeRefused = false;\\n    try {\\n      if (attribute === 'AXValue') await app.setValue('1', 'must not be written');\\n      else await app.selectText('1', 'Fixture');\\n    } catch (error) {\\n      if (!error.message.includes('writability of ' + attribute + ' failed with error -25200')) throw error;\\n      writeRefused = true;\\n    }\\n    if (!writeRefused) throw new Error('Write accepted with unknown writability');\\n    results.push({ attribute, unknown: true, otherNodesRetained: true, writeRefused });\\n  } finally { await unlink('fixture-ax-writability-failure'); }\\n}\\ninspect(results);\",\"description\":\"Retain AX trees with unknown writability and reject writes\"}"}],"source":{"kind":"model","provider":"deepseek-official","model":"deepseek-v4-flash-vision-exp"},"id":"{{message:8}}"},"usage":{"inputTokens":3,"outputTokens":3},"stream":[{"type":"chunk","time":1794226988058,"chunk":{"type":"block-start","index":0,"blockType":"tool-call"}},{"type":"chunk","time":1794226988058,"chunk":{"type":"block-end","index":0,"block":{"type":"tool-call","id":"cua-ptc-writability","name":"run_code","arguments":"{\"code\":\"const { writeFile, unlink } = await import('node:fs/promises');\\nconst { resolve } = await import('node:path');\\nconst { pathToFileURL } = await import('node:url');\\nconst { connectCua } = await import(pathToFileURL(resolve('cua-bridge.mjs')).href);\\nconst cua = connectCua(() => tools.cua);\\nconst app = await cua.getApp('test.fixture');\\nconst results = [];\\nfor (const attribute of ['AXValue', 'AXSelectedTextRange']) {\\n  await writeFile('fixture-ax-writability-failure', attribute);\\n  try {\\n    const tree = await app.getAccessibilityTree();\\n    const field = attribute === 'AXValue' ? 'valueWritable' : 'selectionWritable';\\n    if (!tree.includes(field + '=unknown') || tree.includes(field + '=true')) throw new Error('Unknown writability was lost');\\n    if (!tree.includes('Increment') || !tree.includes('Count:')) throw new Error('Other nodes were lost');\\n    let writeRefused = false;\\n    try {\\n      if (attribute === 'AXValue') await app.setValue('1', 'must not be written');\\n      else await app.selectText('1', 'Fixture');\\n    } catch (error) {\\n      if (!error.message.includes('writability of ' + attribute + ' failed with error -25200')) throw error;\\n      writeRefused = true;\\n    }\\n    if (!writeRefused) throw new Error('Write accepted with unknown writability');\\n    results.push({ attribute, unknown: true, otherNodesRetained: true, writeRefused });\\n  } finally { await unlink('fixture-ax-writability-failure'); }\\n}\\ninspect(results);\",\"description\":\"Retain AX trees with unknown writability and reject writes\"}"}}},{"type":"chunk","time":1794226988058,"chunk":{"type":"usage","usage":{"inputTokens":3,"outputTokens":3}}},{"type":"chunk","time":1794226988058,"chunk":{"type":"finish","reason":{"kind":"tool-calls"}}}]},"surfaceOp":"append"}
@@ -60,7 +60,7 @@
 {"type":"tool/ptc-dispatch","data":{"rootCallId":"cua-ptc-writability","parentCallId":"cua-ptc-writability","subCallId":"cua-ptc-writability:ptc:5","name":"cua","arguments":{"method":"getState","params":{"appHandle":"fixture-app"}},"isError":false,"content":[{"type":"text","text":"{\"appHandle\":\"fixture-app\",\"windowId\":null,\"focusedWindowId\":7,\"snapshotId\":\"fixture-snapshot-4\",\"tree\":\"AX scope=application offsets=UTF-16\\n- [0] AXButton \\\"Increment\\\"\\n- [1] AXTextField valueWritable=true selectionWritable=unknown value=\\\"\\\" placeholder=\\\"Fixture text\\\"\\n- AXStaticText \\\"Count: 1\\\"\\n- [2] AXMenuBarItem \\\"File\\\"\\nMenu contents omitted; request includeMenus=true to expand.\",\"windows\":[{\"windowId\":7,\"appName\":\"Fixture\",\"title\":\"CUA fixture\",\"bounds\":{\"x\":40,\"y\":60,\"width\":320,\"height\":200},\"isOnScreen\":true,\"onCurrentSpace\":true},{\"windowId\":8,\"appName\":\"Fixture\",\"title\":\"AXless canvas\",\"bounds\":{\"x\":40,\"y\":60,\"width\":320,\"height\":200},\"isOnScreen\":true,\"onCurrentSpace\":true},{\"windowId\":9,\"appName\":\"Fixture\",\"title\":\"Transformed window\",\"bounds\":{\"x\":16,\"y\":512,\"width\":131,\"height\":145},\"isOnScreen\":true,\"onCurrentSpace\":true}],\"truncated\":false}"}]}}
 {"type":"tool/ptc-dispatch-start","data":{"rootCallId":"cua-ptc-writability","parentCallId":"cua-ptc-writability","subCallId":"cua-ptc-writability:ptc:6","name":"cua","arguments":{"method":"selectText","params":{"appHandle":"fixture-app","snapshotId":"fixture-snapshot-4","elementId":"1","text":"Fixture"}}}}
 {"type":"tool/ptc-dispatch","data":{"rootCallId":"cua-ptc-writability","parentCallId":"cua-ptc-writability","subCallId":"cua-ptc-writability:ptc:6","name":"cua","arguments":{"method":"selectText","params":{"appHandle":"fixture-app","snapshotId":"fixture-snapshot-4","elementId":"1","text":"Fixture"}},"isError":true,"content":[{"type":"text","text":"Error: NATIVE_ERROR: Accessibility read writability of AXSelectedTextRange failed with error -25200"}]}}
-{"type":"tool/result","data":{"turn":1,"step":3,"message":{"role":"tool","source":{"kind":"tool","callId":"cua-ptc-writability"},"toolCallId":"cua-ptc-writability","content":[{"type":"text","text":"[\n  {\n    \"attribute\": \"AXValue\",\n    \"unknown\": true,\n    \"otherNodesRetained\": true,\n    \"writeRefused\": true\n  },\n  {\n    \"attribute\": \"AXSelectedTextRange\",\n    \"unknown\": true,\n    \"otherNodesRetained\": true,\n    \"writeRefused\": true\n  }\n]"}],"isError":false,"id":"{{message:9}}"}},"sourceEventSeqs":[48],"surfaceOp":"append"}
+{"type":"tool/result","data":{"turn":1,"step":3,"message":{"role":"tool","source":{"kind":"tool","callId":"cua-ptc-writability"},"toolCallId":"cua-ptc-writability","content":[{"type":"text","text":"[\n  {\n    attribute: 'AXValue',\n    unknown: true,\n    otherNodesRetained: true,\n    writeRefused: true\n  },\n  {\n    attribute: 'AXSelectedTextRange',\n    unknown: true,\n    otherNodesRetained: true,\n    writeRefused: true\n  }\n]"}],"isError":false,"id":"{{message:9}}"}},"sourceEventSeqs":[48],"surfaceOp":"append"}
 {"type":"step/end","data":{"turn":1,"step":3}}
 {"type":"step/start","data":{"turn":1,"step":4}}
 {"type":"assistant/message","data":{"turn":1,"step":4,"message":{"role":"assistant","content":[{"type":"text","text":"Received native CUA values and PNG screenshots through PTC. The object API captured a 90-byte PNG after two AX read failures and refused element input without an accessibility snapshot. Both writability probe failures left the tree readable with unknown fields and kept writes strict."}],"source":{"kind":"model","provider":"deepseek-official","model":"deepseek-v4-flash-vision-exp"},"id":"{{message:10}}"},"usage":{"inputTokens":3,"outputTokens":3},"stream":[{"type":"chunk","time":1794226988201,"chunk":{"type":"block-start","index":0,"blockType":"text"}},{"type":"chunk","time":1794226988201,"chunk":{"type":"block-end","index":0,"block":{"type":"text","text":"Received native CUA values and PNG screenshots through PTC. The object API captured a 90-byte PNG after two AX read failures and refused element input without an accessibility snapshot. Both writability probe failures left the tree readable with unknown fields and kept writes strict."}}},{"type":"chunk","time":1794226988201,"chunk":{"type":"usage","usage":{"inputTokens":3,"outputTokens":3}}},{"type":"chunk","time":1794226988201,"chunk":{"type":"finish","reason":{"kind":"stop"}}}]},"surfaceOp":"append"}

snapshots/session/cua-ptc/system-prompt.expected.md

@@ -102,7 +102,7 @@ Inside the program:
 - Call tools as `await tools.name(args)` — quoted access for exotic names: `tools["my-tool"](args)`. Every call resolves to its declared value, which may be an Image. Tool arguments must be lossless JSON.
 - A FAILED tool call rejects with `ToolCallError`, whose `toolName` identifies the failed tool and whose `message` is human-readable — `try/catch` it to handle and continue.
 - Independent read-only calls MAY overlap under `Promise.all` (safe calls run concurrently; mutating calls run alone, in submission order). Sequence dependent work with `await`.
-- Show selected results with `inspect(...values)`. Each call accepts JSON values and Images, including Images nested in arrays and objects. Images expose a readable file path and its pixel dimensions. Extract just what you need.
+- Show selected values and images with `inspect(...values)`; every other intermediate result stays out of the conversation, so extract just what you need.
 
 Program-only SDK bindings:
 

snapshots/session/mcp-resources-ptc/system-prompt.expected.md

@@ -54,7 +54,7 @@ Inside the program:
 - Call tools as `await tools.name(args)` — quoted access for exotic names: `tools["my-tool"](args)`. Every call resolves to its declared value, which may be an Image. Tool arguments must be lossless JSON.
 - A FAILED tool call rejects with `ToolCallError`, whose `toolName` identifies the failed tool and whose `message` is human-readable — `try/catch` it to handle and continue.
 - Independent read-only calls MAY overlap under `Promise.all` (safe calls run concurrently; mutating calls run alone, in submission order). Sequence dependent work with `await`.
-- Show selected results with `inspect(...values)`. Each call accepts JSON values and Images, including Images nested in arrays and objects. Images expose a readable file path and its pixel dimensions. Extract just what you need.
+- Show selected values and images with `inspect(...values)`; every other intermediate result stays out of the conversation, so extract just what you need.
 
 Program-only SDK bindings:
 

snapshots/session/office-progressive-ptc/session.v4.jsonl

@@ -17,7 +17,7 @@
 {"type":"tool/call","data":{"turn":1,"step":1,"callId":"call_office_skill","name":"run_code","arguments":"{\"code\":\"inspect(await tools.skill({\\\"name\\\":\\\"office\\\"}));\",\"description\":\"Load task instructions\"}"}}
 {"type":"tool/ptc-dispatch-start","data":{"rootCallId":"call_office_skill","parentCallId":"call_office_skill","subCallId":"call_office_skill:ptc:1","name":"skill","arguments":{"name":"office"}}}
 {"type":"tool/ptc-dispatch","data":{"rootCallId":"call_office_skill","parentCallId":"call_office_skill","subCallId":"call_office_skill:ptc:1","name":"skill","arguments":{"name":"office"},"isError":false,"content":[{"type":"text","text":"<skill_content name=\"office\">\n<skill_resources>\nBase directory for this skill: {{cwd}}/office-skills/office\nResolve relative paths mentioned by this skill against the base directory before using them. Load referenced resources only as needed.\n</skill_resources>\n\n<skill_instructions>\n# Office task routing\n\nLoad `office-docx`, `office-pptx`, or `office-xlsx` to create, read, or edit; operations alone need no format skill.\n\nIf `office_render`, `office_convert`, or `office_recalculate` is not in your tools, load `office` with `skill`; they appear next step.\n\n- `office_render`: Office or PDF images, visual checks, and appearance or slide checks required by the format skill; skip it when asked.\n- `office_recalculate`: formulas, with `format: \"xlsx\"` or `\"ods\"`.\n- `office_convert` with `format`; Office inputs only, not PDFs. For PDF export alone, check the copy is nonempty and starts with `%PDF-`, then deliver it without rendering.\n\nCopy temporary outputs to the destination, keeping existing files unless asked to replace them. Reuse loaded skills.\n</skill_instructions>\n</skill_content>"}]}}
-{"type":"tool/result","data":{"turn":1,"step":1,"message":{"role":"tool","source":{"kind":"tool","callId":"call_office_skill"},"toolCallId":"call_office_skill","content":[{"type":"text","text":"{\n  \"name\": \"office\",\n  \"provider\": \"dsh-office\",\n  \"resourceBase\": {\n    \"kind\": \"directory\",\n    \"path\": \"{{cwd}}/office-skills/office\"\n  },\n  \"content\": \"# Office task routing\\n\\nLoad `office-docx`, `office-pptx`, or `office-xlsx` to create, read, or edit; operations alone need no format skill.\\n\\nIf `office_render`, `office_convert`, or `office_recalculate` is not in your tools, load `office` with `skill`; they appear next step.\\n\\n- `office_render`: Office or PDF images, visual checks, and appearance or slide checks required by the format skill; skip it when asked.\\n- `office_recalculate`: formulas, with `format: \\\"xlsx\\\"` or `\\\"ods\\\"`.\\n- `office_convert` with `format`; Office inputs only, not PDFs. For PDF export alone, check the copy is nonempty and starts with `%PDF-`, then deliver it without rendering.\\n\\nCopy temporary outputs to the destination, keeping existing files unless asked to replace them. Reuse loaded skills.\"\n}"}],"isError":false,"id":"{{message:6}}"}},"sourceEventSeqs":[15],"surfaceOp":"append"}
+{"type":"tool/result","data":{"turn":1,"step":1,"message":{"role":"tool","source":{"kind":"tool","callId":"call_office_skill"},"toolCallId":"call_office_skill","content":[{"type":"text","text":"{\n  name: 'office',\n  provider: 'dsh-office',\n  resourceBase: {\n    kind: 'directory',\n    path: '{{cwd}}/office-skills/office'\n  },\n  content: '# Office task routing\\n' +\n    '\\n' +\n    'Load `office-docx`, `office-pptx`, or `office-xlsx` to create, read, or edit; operations alone need no format skill.\\n' +\n    '\\n' +\n    'If `office_render`, `office_convert`, or `office_recalculate` is not in your tools, load `office` with `skill`; they appear next step.\\n' +\n    '\\n' +\n    '- `office_render`: Office or PDF images, visual checks, and appearance or slide checks required by the format skill; skip it when asked.\\n' +\n    '- `office_recalculate`: formulas, with `format: \"xlsx\"` or `\"ods\"`.\\n' +\n    '- `office_convert` with `format`; Office inputs only, not PDFs. For PDF export alone, check the copy is nonempty and starts with `%PDF-`, then deliver it without rendering.\\n' +\n    '\\n' +\n    'Copy temporary outputs to the destination, keeping existing files unless asked to replace them. Reuse loaded skills.'\n}"}],"isError":false,"id":"{{message:6}}"}},"sourceEventSeqs":[15],"surfaceOp":"append"}
 {"type":"step/end","data":{"turn":1,"step":1}}
 {"type":"step/start","data":{"turn":1,"step":2}}
 {"type":"system/message","data":{"turn":1,"step":2,"message":{"role":"system","content":[{"type":"text","text":"{{system}}"}],"source":{"kind":"system-prompt"},"id":"{{message:7}}"}},"sourceEventSeqs":[7],"surfaceOp":{"op":"replace","startSeq":7,"endSeq":7}}
@@ -26,7 +26,7 @@
 {"type":"tool/call","data":{"turn":1,"step":2,"callId":"call_office_2","name":"run_code","arguments":"{\"code\":\"inspect(await tools.skill({\\\"name\\\":\\\"office-docx\\\"}));\",\"description\":\"Load task instructions\"}"}}
 {"type":"tool/ptc-dispatch-start","data":{"rootCallId":"call_office_2","parentCallId":"call_office_2","subCallId":"call_office_2:ptc:1","name":"skill","arguments":{"name":"office-docx"}}}
 {"type":"tool/ptc-dispatch","data":{"rootCallId":"call_office_2","parentCallId":"call_office_2","subCallId":"call_office_2:ptc:1","name":"skill","arguments":{"name":"office-docx"},"isError":false,"content":[{"type":"text","text":"<skill_content name=\"office-docx\">\n<skill_resources>\nBase directory for this skill: {{cwd}}/office-skills/office-docx\nResolve relative paths mentioned by this skill against the base directory before using them. Load referenced resources only as needed.\n</skill_resources>\n\n<skill_instructions>\n# Word documents\n\nUse `python-docx` for DOCX creation and ordinary edits unless an explicit user or applicable AGENTS.md requirement names another library. Run Python as described in the Python environment section below.\n\nKeep scripts, intermediate files, and final documents in the workspace; runtime and skill resources are read-only. Use the requested language and preserve the document's design unless a redesign is requested.\n\nName Python scripts for their task, such as `edit_docx.py`; avoid module names such as `inspect.py`, `docx.py`, or `zipfile.py` that shadow imported libraries.\n\n## Create and edit\n\nIf `office_render`, `office_convert`, or `office_recalculate` is not in your tools, load `office` with `skill`; they appear next step.\n\npython-docx cannot read legacy `.doc` files. Use `office_convert` with `format: \"docx\"`, copy the result into the workspace, then edit the copy.\n\nFor text edits, locate the target by its text or position and inspect only its runs and formatting. A title can be an ordinary paragraph without a Title style or document-property title. Update the relevant runs; assigning paragraph `.text` destroys run formatting. Save a new file unless an in-place edit is requested. For ordinary text edits, do not unzip the document, enumerate ZIP members, dump raw XML, or inspect core properties, headers, or footers unrelated to the target. Investigate these only for a concrete failure, an ambiguous target, an unsupported feature, or an explicit user requirement. Rebuilding a document can lose unsupported features.\n\nUse paragraph styles for headings and body text. Size tables for the section that contains them, and account for merged cells and nested tables. Chinese, Japanese, and Korean text may need an explicit `w:eastAsia` font assignment in addition to `run.font.name`; font names alone do not establish glyph availability or rendered appearance.\n\nFor a new document, use the selected Python executable:\n\n```python\nfrom docx import Document\n\ndocument = Document()\ndocument.add_heading(\"Project report\", level=0)\ndocument.add_paragraph(\"Summary\", style=\"Heading 1\")\ndocument.add_paragraph(\"The requested findings go here.\")\ndocument.save(\"report.docx\")\n```\n\npython-docx does not paginate or render documents. Do not represent ordinary replacement, colored text, or comments as tracked changes. When real revisions or unsupported OOXML features matter, preserve their package parts and verify the requested operation rather than silently discarding them. When writing package parts directly, create a new ZIP and copy each unchanged member once; appending a replacement member duplicates its name and corrupts the package.\n\n## Check and deliver\n\nReopen the saved document with python-docx and verify the requested content, related formatting, and unchanged content that must be preserved. These checks complete routine validation; deliver the file once they pass. ZIP integrity, package inventories, and metadata matter only for a concrete failure.\n\nCall `office_render` on the changed page when the user asks for images or visual inspection, or when styling, font color, spacing, sizing, or text alignment is the requested edit; a request to skip rendering takes precedence. New documents and content edits, with incidental basic formatting, are checked with python-docx alone. For PDF output, call `office_convert` with `format: \"pdf\"`, then copy the returned temporary file to the destination.\n\nA fully transparent or uniformly blank image of populated content means the renderer failed, which ends visual checks for the task and leaves the source as it is. When content and structure checks pass, state the visual status only as \"Content and structure checks passed; visual inspection was not completed.\" or \"内容与结构检查通过；视觉检查未完成。\".\n\nCall `present({\"files\":[{\"path\":\"report.docx\"}]})` with the actual final DOCX path. It exposes the current file without copying or preserving its bytes; keep it in place and omit temporary QA reports unless requested. If `present` is unavailable, deliver the final workspace path with the session's supported method.\n\n## Python environment\n\nFollow an explicit user or applicable AGENTS.md requirement for a Python environment. Otherwise call `load_workspace_dependencies` and run scripts with its Python executable and bundled libraries. Do not install packages or search for a sysLine truncated
-{"type":"tool/result","data":{"turn":1,"step":2,"message":{"role":"tool","source":{"kind":"tool","callId":"call_office_2"},"toolCallId":"call_office_2","content":[{"type":"text","text":"{\n  \"name\": \"office-docx\",\n  \"provider\": \"dsh-office\",\n  \"resourceBase\": {\n    \"kind\": \"directory\",\n    \"path\": \"{{cwd}}/office-skills/office-docx\"\n  },\n  \"content\": \"# Word documents\\n\\nUse `python-docx` for DOCX creation and ordinary edits unless an explicit user or applicable AGENTS.md requirement names another library. Run Python as described in the Python environment section below.\\n\\nKeep scripts, intermediate files, and final documents in the workspace; runtime and skill resources are read-only. Use the requested language and preserve the document's design unless a redesign is requested.\\n\\nName Python scripts for their task, such as `edit_docx.py`; avoid module names such as `inspect.py`, `docx.py`, or `zipfile.py` that shadow imported libraries.\\n\\n## Create and edit\\n\\nIf `office_render`, `office_convert`, or `office_recalculate` is not in your tools, load `office` with `skill`; they appear next step.\\n\\npython-docx cannot read legacy `.doc` files. Use `office_convert` with `format: \\\"docx\\\"`, copy the result into the workspace, then edit the copy.\\n\\nFor text edits, locate the target by its text or position and inspect only its runs and formatting. A title can be an ordinary paragraph without a Title style or document-property title. Update the relevant runs; assigning paragraph `.text` destroys run formatting. Save a new file unless an in-place edit is requested. For ordinary text edits, do not unzip the document, enumerate ZIP members, dump raw XML, or inspect core properties, headers, or footers unrelated to the target. Investigate these only for a concrete failure, an ambiguous target, an unsupported feature, or an explicit user requirement. Rebuilding a document can lose unsupported features.\\n\\nUse paragraph styles for headings and body text. Size tables for the section that contains them, and account for merged cells and nested tables. Chinese, Japanese, and Korean text may need an explicit `w:eastAsia` font assignment in addition to `run.font.name`; font names alone do not establish glyph availability or rendered appearance.\\n\\nFor a new document, use the selected Python executable:\\n\\n```python\\nfrom docx import Document\\n\\ndocument = Document()\\ndocument.add_heading(\\\"Project report\\\", level=0)\\ndocument.add_paragraph(\\\"Summary\\\", style=\\\"Heading 1\\\")\\ndocument.add_paragraph(\\\"The requested findings go here.\\\")\\ndocument.save(\\\"report.docx\\\")\\n```\\n\\npython-docx does not paginate or render documents. Do not represent ordinary replacement, colored text, or comments as tracked changes. When real revisions or unsupported OOXML features matter, preserve their package parts and verify the requested operation rather than silently discarding them. When writing package parts directly, create a new ZIP and copy each unchanged member once; appending a replacement member duplicates its name and corrupts the package.\\n\\n## Check and deliver\\n\\nReopen the saved document with python-docx and verify the requested content, related formatting, and unchanged content that must be preserved. These checks complete routine validation; deliver the file once they pass. ZIP integrity, package inventories, and metadata matter only for a concrete failure.\\n\\nCall `office_render` on the changed page when the user asks for images or visual inspection, or when styling, font color, spacing, sizing, or text alignment is the requested edit; a request to skip rendering takes precedence. New documents and content edits, with incidental basic formatting, are checked with python-docx alone. For PDF output, call `office_convert` with `format: \\\"pdf\\\"`, then copy the returned temporary file to the destination.\\n\\nA fully transparent or uniformly blank image of populated content means the renderer failed, which ends visual checks for the task and leaves the source as it is. When content and structure checks pass, state the visual status only as \\\"Content and structure checks passed; visual inspection was not completed.\\\" or \\\"内容与结构检查通过；视觉检查未完成。\\\".\\n\\nCall `present({\\\"files\\\":[{\\\"path\\\":\\\"report.docx\\\"}]})` with the actual final DOCX path. It exposes the current file without copying or preserving its bytes; keep it in place and omit temporary QA reports unless requested. If `present` is unavailable, deliver the final workspace path with the session's supported method.\\n\\n## Python environment\\n\\nFollow an explicit user or applicable AGENTS.md requirement for a Python environment. Otherwise call `load_workspace_dependencies` and run scripts with its Python executable and bundled libraries. Do not install packages or search for a system Python for the default workflow. If the tool is unavailaLine truncated
+{"type":"tool/result","data":{"turn":1,"step":2,"message":{"role":"tool","source":{"kind":"tool","callId":"call_office_2"},"toolCallId":"call_office_2","content":[{"type":"text","text":"{\n  name: 'office-docx',\n  provider: 'dsh-office',\n  resourceBase: {\n    kind: 'directory',\n    path: '{{cwd}}/office-skills/office-docx'\n  },\n  content: '# Word documents\\n' +\n    '\\n' +\n    'Use `python-docx` for DOCX creation and ordinary edits unless an explicit user or applicable AGENTS.md requirement names another library. Run Python as described in the Python environment section below.\\n' +\n    '\\n' +\n    \"Keep scripts, intermediate files, and final documents in the workspace; runtime and skill resources are read-only. Use the requested language and preserve the document's design unless a redesign is requested.\\n\" +\n    '\\n' +\n    'Name Python scripts for their task, such as `edit_docx.py`; avoid module names such as `inspect.py`, `docx.py`, or `zipfile.py` that shadow imported libraries.\\n' +\n    '\\n' +\n    '## Create and edit\\n' +\n    '\\n' +\n    'If `office_render`, `office_convert`, or `office_recalculate` is not in your tools, load `office` with `skill`; they appear next step.\\n' +\n    '\\n' +\n    'python-docx cannot read legacy `.doc` files. Use `office_convert` with `format: \"docx\"`, copy the result into the workspace, then edit the copy.\\n' +\n    '\\n' +\n    'For text edits, locate the target by its text or position and inspect only its runs and formatting. A title can be an ordinary paragraph without a Title style or document-property title. Update the relevant runs; assigning paragraph `.text` destroys run formatting. Save a new file unless an in-place edit is requested. For ordinary text edits, do not unzip the document, enumerate ZIP members, dump raw XML, or inspect core properties, headers, or footers unrelated to the target. Investigate these only for a concrete failure, an ambiguous target, an unsupported feature, or an explicit user requirement. Rebuilding a document can lose unsupported features.\\n' +\n    '\\n' +\n    'Use paragraph styles for headings and body text. Size tables for the section that contains them, and account for merged cells and nested tables. Chinese, Japanese, and Korean text may need an explicit `w:eastAsia` font assignment in addition to `run.font.name`; font names alone do not establish glyph availability or rendered appearance.\\n' +\n    '\\n' +\n    'For a new document, use the selected Python executable:\\n' +\n    '\\n' +\n    '```python\\n' +\n    'from docx import Document\\n' +\n    '\\n' +\n    'document = Document()\\n' +\n    'document.add_heading(\"Project report\", level=0)\\n' +\n    'document.add_paragraph(\"Summary\", style=\"Heading 1\")\\n' +\n    'document.add_paragraph(\"The requested findings go here.\")\\n' +\n    'document.save(\"report.docx\")\\n' +\n    '```\\n' +\n    '\\n' +\n    'python-docx does not paginate or render documents. Do not represent ordinary replacement, colored text, or comments as tracked changes. When real revisions or unsupported OOXML features matter, preserve their package parts and verify the requested operation rather than silently discarding them. When writing package parts directly, create a new ZIP and copy each unchanged member once; appending a replacement member duplicates its name and corrupts the package.\\n' +\n    '\\n' +\n    '## Check and deliver\\n' +\n    '\\n' +\n    'Reopen the saved document with python-docx and verify the requested content, related formatting, and unchanged content that must be preserved. These checks complete routine validation; deliver the file once they pass. ZIP integrity, package inventories, and metadata matter only for a concrete failure.\\n' +\n    '\\n' +\n    'Call `office_render` on the changed page when the user asks for images or visual inspection, or when styling, font color, spacing, sizing, or text alignment is the requested edit; a request to skip rendering takes precedence. New documents and content edits, with incidental basic formatting, are checked with python-docx alone. For PDF output, call `office_convert` with `format: \"pdf\"`, then copy the returned temporary file to the destination.\\n' +\n    '\\n' +\n    'A fully transparent or uniformly blank image of populated content means the renderer failed, which ends visual checks for the task and leaves the source as it is. When content and structure checks pass, state the visual status only as \"Content and structure checks passed; visual inspection was not completed.\" or \"内容与结构检查通过；视觉检查未完成。\".\\n' +\n    '\\n' +\n    'Call `present({\"files\":[{\"path\":\"report.docx\"}]})` with the actual final DOCX path. It exposes the current file without copying or preserving its bytes; keep it in place and omit temporary QA reports unless requested. If `present` is unavailable, deliver the final workspace path with the session\\'s supported method.\\n' +\n  Line truncated
 {"type":"step/end","data":{"turn":1,"step":2}}
 {"type":"step/start","data":{"turn":1,"step":3}}
 {"type":"assistant/message","data":{"turn":1,"step":3,"message":{"role":"assistant","content":[{"type":"text","text":"DONE"}],"source":{"kind":"model","provider":"deepseek-official","model":"deepseek-v4-flash"},"id":"{{message:10}}"},"usage":{"inputTokens":12,"outputTokens":2,"cacheReadTokens":10},"stream":[{"type":"chunk","time":1794195777349,"chunk":{"type":"block-start","index":0,"blockType":"text"}},{"type":"chunk","time":1794195777349,"chunk":{"type":"block-end","index":0,"block":{"type":"text","text":"DONE"}}},{"type":"chunk","time":1794195777349,"chunk":{"type":"usage","usage":{"inputTokens":12,"outputTokens":2,"cacheReadTokens":10}}},{"type":"chunk","time":1794195777349,"chunk":{"type":"finish","reason":{"kind":"stop"}}}]},"surfaceOp":"append"}

snapshots/session/office-progressive-ptc/system-prompt.expected.md

@@ -42,7 +42,7 @@ Inside the program:
 - Call tools as `await tools.name(args)` — quoted access for exotic names: `tools["my-tool"](args)`. Every call resolves to its declared value, which may be an Image. Tool arguments must be lossless JSON.
 - A FAILED tool call rejects with `ToolCallError`, whose `toolName` identifies the failed tool and whose `message` is human-readable — `try/catch` it to handle and continue.
 - Independent read-only calls MAY overlap under `Promise.all` (safe calls run concurrently; mutating calls run alone, in submission order). Sequence dependent work with `await`.
-- Show selected results with `inspect(...values)`. Each call accepts JSON values and Images, including Images nested in arrays and objects. Images expose a readable file path and its pixel dimensions. Extract just what you need.
+- Show selected values and images with `inspect(...values)`; every other intermediate result stays out of the conversation, so extract just what you need.
 
 Program-only SDK bindings:
 
@@ -555,7 +555,7 @@ Inside the program:
 - Call tools as `await tools.name(args)` — quoted access for exotic names: `tools["my-tool"](args)`. Every call resolves to its declared value, which may be an Image. Tool arguments must be lossless JSON.
 - A FAILED tool call rejects with `ToolCallError`, whose `toolName` identifies the failed tool and whose `message` is human-readable — `try/catch` it to handle and continue.
 - Independent read-only calls MAY overlap under `Promise.all` (safe calls run concurrently; mutating calls run alone, in submission order). Sequence dependent work with `await`.
-- Show selected results with `inspect(...values)`. Each call accepts JSON values and Images, including Images nested in arrays and objects. Images expose a readable file path and its pixel dimensions. Extract just what you need.
+- Show selected values and images with `inspect(...values)`; every other intermediate result stays out of the conversation, so extract just what you need.
 
 Program-only SDK bindings:
 

snapshots/session/office-render-tools-ptc/writer.expected.jsonl

@@ -14,30 +14,30 @@
 {"type":"request/header","data":{"header":{"config":{"provider":"deepseek-official","model":"deepseek-v4-flash"},"tools":"{{tools}}"},"reason":"initial"}}
 {"type":"request/context","data":{"provider":"deepseek-official","model":"deepseek-v4-flash"}}
 {"type":"session/title","data":{"title":"/office-docx Load the Office rendering","messageSeqs":[8],"source":{"kind":"fallback"}}}
-{"type":"assistant/message","data":{"turn":1,"step":1,"message":{"role":"assistant","content":[{"type":"tool-call","id":"call_office_skill","name":"run_code","arguments":"{\"code\":\"inspect(await tools.skill({\\\"name\\\":\\\"office-xlsx\\\"}));\",\"description\":\"Use the Office workflow\"}"}],"source":{"kind":"model","provider":"deepseek-official","model":"deepseek-v4-flash"},"id":"{{message:6}}"},"stream":[{"type":"chunk","time":1794204018722,"chunk":{"type":"block-start","index":0,"blockType":"tool-call"}},{"type":"chunk","time":1794204018722,"chunk":{"type":"block-end","index":0,"block":{"type":"tool-call","id":"call_office_skill","name":"run_code","arguments":"{\"code\":\"inspect(await tools.skill({\\\"name\\\":\\\"office-xlsx\\\"}));\",\"description\":\"Use the Office workflow\"}"}}},{"type":"chunk","time":1794204018722,"chunk":{"type":"finish","reason":{"kind":"tool-calls"}}}]},"surfaceOp":"append"}
+{"type":"assistant/message","data":{"turn":1,"step":1,"message":{"role":"assistant","content":[{"type":"tool-call","id":"call_office_skill","name":"run_code","arguments":"{\"code\":\"inspect(await tools.skill({\\\"name\\\":\\\"office-xlsx\\\"}));\",\"description\":\"Use the Office workflow\"}"}],"source":{"kind":"model","provider":"deepseek-official","model":"deepseek-v4-flash"},"id":"{{message:6}}"},"stream":[{"type":"chunk","time":1794283520828,"chunk":{"type":"block-start","index":0,"blockType":"tool-call"}},{"type":"chunk","time":1794283520828,"chunk":{"type":"block-end","index":0,"block":{"type":"tool-call","id":"call_office_skill","name":"run_code","arguments":"{\"code\":\"inspect(await tools.skill({\\\"name\\\":\\\"office-xlsx\\\"}));\",\"description\":\"Use the Office workflow\"}"}}},{"type":"chunk","time":1794283520828,"chunk":{"type":"finish","reason":{"kind":"tool-calls"}}}]},"surfaceOp":"append"}
 {"type":"tool/call","data":{"turn":1,"step":1,"callId":"call_office_skill","name":"run_code","arguments":"{\"code\":\"inspect(await tools.skill({\\\"name\\\":\\\"office-xlsx\\\"}));\",\"description\":\"Use the Office workflow\"}"}}
 {"type":"tool/ptc-dispatch-start","data":{"rootCallId":"call_office_skill","parentCallId":"call_office_skill","subCallId":"call_office_skill:ptc:1","name":"skill","arguments":{"name":"office-xlsx"}}}
 {"type":"tool/ptc-dispatch","data":{"rootCallId":"call_office_skill","parentCallId":"call_office_skill","subCallId":"call_office_skill:ptc:1","name":"skill","arguments":{"name":"office-xlsx"},"isError":false,"content":[{"type":"text","text":"<skill_content name=\"office-xlsx\">\n<skill_resources>\nBase directory for this skill: {{cwd}}/office-skills/office-xlsx\nResolve relative paths mentioned by this skill against the base directory before using them. Load referenced resources only as needed.\n</skill_resources>\n\n<skill_instructions>\n# Excel workbooks\n\nUse `openpyxl` and `pandas` unless an explicit user or applicable AGENTS.md requirement names another library. Run Python as described in the Python environment section below.\n\nKeep scripts, working files, and outputs in the task workspace. The runtime and skill directory are shared read-only resources. Save to a new workbook unless the user requests an in-place edit.\n\nName Python scripts for their task, such as `edit_xlsx.py`; avoid module names such as `inspect.py`, `openpyxl.py`, or `pandas.py` that shadow imported libraries.\n\n## Read and modify\n\nIf `office_render`, `office_convert`, or `office_recalculate` is not in your tools, load `office` with `skill`; they appear next step.\n\nopenpyxl cannot read legacy `.xls` files. Use `office_convert` with `format: \"xlsx\"`, copy the result into the workspace, then edit the copy.\n\nUse `openpyxl` for existing XLSX workbooks and targeted cell or style changes. Load with `data_only=False` when formulas must survive. Inspect sheet names, the affected cell types, formulas, styles, merged ranges, and referenced ranges before editing. Check the saved file by reopening it, including unchanged content that the request requires preserving.\n\nUse pandas for data analysis and transformations. A DataFrame is not the workbook: exporting it over an existing file can lose sheets, formulas, charts, and formatting. Write analysis results back to the intended ranges with openpyxl. `XlsxWriter` can create new workbooks, but cannot read or modify existing ones.\n\nFor a new workbook, the selected Python environment can write editable values, styles, and formulas directly:\n\n```python\nfrom openpyxl import Workbook\nfrom openpyxl.styles import Font\n\nworkbook = Workbook()\nsheet = workbook.active\nsheet.title = \"Revenue\"\nfor row in [(\"Quarter\", \"Revenue\"), (\"Q1\", 12), (\"Q2\", 18), (\"Total\", \"=SUM(B2:B3)\")]:\n    sheet.append(row)\nfor cell in sheet[1]:\n    cell.font = Font(bold=True)\nsheet.column_dimensions[\"A\"].width = 18\nsheet.column_dimensions[\"B\"].width = 18\nworkbook.save(\"report.xlsx\")\n```\n\nPreserve numbers, dates, booleans, and identifiers as the intended cell types; formatting is not a type conversion. For modifications, load the existing file with `openpyxl.load_workbook(\"input.xlsx\", data_only=False)`, change the requested ranges, and save a separate result. Avoid a DataFrame round trip when workbook features must survive.\n\nWriting a formula does not calculate its result. openpyxl and XlsxWriter do not evaluate formulas; `data_only=True` returns cached values that may be absent or stale. When calculated results must be refreshed, call `office_recalculate` with `format: \"xlsx\"`, and copy the returned temporary workbook to a new file in the workspace. Open the copy with openpyxl using both `data_only=False` and `data_only=True` to check that formulas survive and cached values match the expected results. If recalculation fails, preserve the original workbook and report that calculated values were not refreshed. Do not replace requested formulas with constants or report stale or absent cached values as newly calculated results.\n\nDo not rename `.xls`, `.xlsb`, encrypted files, or macro-enabled files to `.xlsx`. They need an appropriate supported operation. `keep_vba=True` can preserve VBA package content in an XLSM workflow, but does not execute or edit macros. Preserve the original and verify each workbook feature the task depends on.\n\n## Check and deliver\n\nReopen the saved workbook with openpyxl and verify the requested sheet names, values, types, formulas, styles, and totals against the source data, including unchanged content that must be preserved. These checks do not calculate formulas or judge workbook appearance.\n\nCall `office_render` when the user asks for images or visual inspection, or when the task targets appearance, such as cell fill colors, text alignment, spacing, or sizing; a request to skip rendering takes precedence. Do not render new workbooks or data or formula edits merely to verify accompanying bold headers or number formats; check these with openpyxl alone.\n\nRender worksheets by `sheet` and the affected `range`; worksheet numbers are not PDF page numbers. Earlier images of an unchanged source remain current. A fully transparent or uniformly blank image of populated content means the renderer failed, which ends visual checks for the task and leaLine truncated
-{"type":"tool/result","data":{"turn":1,"step":1,"message":{"role":"tool","source":{"kind":"tool","callId":"call_office_skill"},"toolCallId":"call_office_skill","content":[{"type":"text","text":"{\n  \"name\": \"office-xlsx\",\n  \"provider\": \"dsh-office\",\n  \"resourceBase\": {\n    \"kind\": \"directory\",\n    \"path\": \"{{cwd}}/office-skills/office-xlsx\"\n  },\n  \"content\": \"# Excel workbooks\\n\\nUse `openpyxl` and `pandas` unless an explicit user or applicable AGENTS.md requirement names another library. Run Python as described in the Python environment section below.\\n\\nKeep scripts, working files, and outputs in the task workspace. The runtime and skill directory are shared read-only resources. Save to a new workbook unless the user requests an in-place edit.\\n\\nName Python scripts for their task, such as `edit_xlsx.py`; avoid module names such as `inspect.py`, `openpyxl.py`, or `pandas.py` that shadow imported libraries.\\n\\n## Read and modify\\n\\nIf `office_render`, `office_convert`, or `office_recalculate` is not in your tools, load `office` with `skill`; they appear next step.\\n\\nopenpyxl cannot read legacy `.xls` files. Use `office_convert` with `format: \\\"xlsx\\\"`, copy the result into the workspace, then edit the copy.\\n\\nUse `openpyxl` for existing XLSX workbooks and targeted cell or style changes. Load with `data_only=False` when formulas must survive. Inspect sheet names, the affected cell types, formulas, styles, merged ranges, and referenced ranges before editing. Check the saved file by reopening it, including unchanged content that the request requires preserving.\\n\\nUse pandas for data analysis and transformations. A DataFrame is not the workbook: exporting it over an existing file can lose sheets, formulas, charts, and formatting. Write analysis results back to the intended ranges with openpyxl. `XlsxWriter` can create new workbooks, but cannot read or modify existing ones.\\n\\nFor a new workbook, the selected Python environment can write editable values, styles, and formulas directly:\\n\\n```python\\nfrom openpyxl import Workbook\\nfrom openpyxl.styles import Font\\n\\nworkbook = Workbook()\\nsheet = workbook.active\\nsheet.title = \\\"Revenue\\\"\\nfor row in [(\\\"Quarter\\\", \\\"Revenue\\\"), (\\\"Q1\\\", 12), (\\\"Q2\\\", 18), (\\\"Total\\\", \\\"=SUM(B2:B3)\\\")]:\\n    sheet.append(row)\\nfor cell in sheet[1]:\\n    cell.font = Font(bold=True)\\nsheet.column_dimensions[\\\"A\\\"].width = 18\\nsheet.column_dimensions[\\\"B\\\"].width = 18\\nworkbook.save(\\\"report.xlsx\\\")\\n```\\n\\nPreserve numbers, dates, booleans, and identifiers as the intended cell types; formatting is not a type conversion. For modifications, load the existing file with `openpyxl.load_workbook(\\\"input.xlsx\\\", data_only=False)`, change the requested ranges, and save a separate result. Avoid a DataFrame round trip when workbook features must survive.\\n\\nWriting a formula does not calculate its result. openpyxl and XlsxWriter do not evaluate formulas; `data_only=True` returns cached values that may be absent or stale. When calculated results must be refreshed, call `office_recalculate` with `format: \\\"xlsx\\\"`, and copy the returned temporary workbook to a new file in the workspace. Open the copy with openpyxl using both `data_only=False` and `data_only=True` to check that formulas survive and cached values match the expected results. If recalculation fails, preserve the original workbook and report that calculated values were not refreshed. Do not replace requested formulas with constants or report stale or absent cached values as newly calculated results.\\n\\nDo not rename `.xls`, `.xlsb`, encrypted files, or macro-enabled files to `.xlsx`. They need an appropriate supported operation. `keep_vba=True` can preserve VBA package content in an XLSM workflow, but does not execute or edit macros. Preserve the original and verify each workbook feature the task depends on.\\n\\n## Check and deliver\\n\\nReopen the saved workbook with openpyxl and verify the requested sheet names, values, types, formulas, styles, and totals against the source data, including unchanged content that must be preserved. These checks do not calculate formulas or judge workbook appearance.\\n\\nCall `office_render` when the user asks for images or visual inspection, or when the task targets appearance, such as cell fill colors, text alignment, spacing, or sizing; a request to skip rendering takes precedence. Do not render new workbooks or data or formula edits merely to verify accompanying bold headers or number formats; check these with openpyxl alone.\\n\\nRender worksheets by `sheet` and the affected `range`; worksheet numbers are not PDF page numbers. Earlier images of an unchanged source remain current. A fully transparent or uniformly blank image of populated content means the renderer failed, which ends visual checks for the task and leaves the source as it is. When content and structure checLine truncated
+{"type":"tool/result","data":{"turn":1,"step":1,"message":{"role":"tool","source":{"kind":"tool","callId":"call_office_skill"},"toolCallId":"call_office_skill","content":[{"type":"text","text":"{\n  name: 'office-xlsx',\n  provider: 'dsh-office',\n  resourceBase: {\n    kind: 'directory',\n    path: '{{cwd}}/office-skills/office-xlsx'\n  },\n  content: '# Excel workbooks\\n' +\n    '\\n' +\n    'Use `openpyxl` and `pandas` unless an explicit user or applicable AGENTS.md requirement names another library. Run Python as described in the Python environment section below.\\n' +\n    '\\n' +\n    'Keep scripts, working files, and outputs in the task workspace. The runtime and skill directory are shared read-only resources. Save to a new workbook unless the user requests an in-place edit.\\n' +\n    '\\n' +\n    'Name Python scripts for their task, such as `edit_xlsx.py`; avoid module names such as `inspect.py`, `openpyxl.py`, or `pandas.py` that shadow imported libraries.\\n' +\n    '\\n' +\n    '## Read and modify\\n' +\n    '\\n' +\n    'If `office_render`, `office_convert`, or `office_recalculate` is not in your tools, load `office` with `skill`; they appear next step.\\n' +\n    '\\n' +\n    'openpyxl cannot read legacy `.xls` files. Use `office_convert` with `format: \"xlsx\"`, copy the result into the workspace, then edit the copy.\\n' +\n    '\\n' +\n    'Use `openpyxl` for existing XLSX workbooks and targeted cell or style changes. Load with `data_only=False` when formulas must survive. Inspect sheet names, the affected cell types, formulas, styles, merged ranges, and referenced ranges before editing. Check the saved file by reopening it, including unchanged content that the request requires preserving.\\n' +\n    '\\n' +\n    'Use pandas for data analysis and transformations. A DataFrame is not the workbook: exporting it over an existing file can lose sheets, formulas, charts, and formatting. Write analysis results back to the intended ranges with openpyxl. `XlsxWriter` can create new workbooks, but cannot read or modify existing ones.\\n' +\n    '\\n' +\n    'For a new workbook, the selected Python environment can write editable values, styles, and formulas directly:\\n' +\n    '\\n' +\n    '```python\\n' +\n    'from openpyxl import Workbook\\n' +\n    'from openpyxl.styles import Font\\n' +\n    '\\n' +\n    'workbook = Workbook()\\n' +\n    'sheet = workbook.active\\n' +\n    'sheet.title = \"Revenue\"\\n' +\n    'for row in [(\"Quarter\", \"Revenue\"), (\"Q1\", 12), (\"Q2\", 18), (\"Total\", \"=SUM(B2:B3)\")]:\\n' +\n    '    sheet.append(row)\\n' +\n    'for cell in sheet[1]:\\n' +\n    '    cell.font = Font(bold=True)\\n' +\n    'sheet.column_dimensions[\"A\"].width = 18\\n' +\n    'sheet.column_dimensions[\"B\"].width = 18\\n' +\n    'workbook.save(\"report.xlsx\")\\n' +\n    '```\\n' +\n    '\\n' +\n    'Preserve numbers, dates, booleans, and identifiers as the intended cell types; formatting is not a type conversion. For modifications, load the existing file with `openpyxl.load_workbook(\"input.xlsx\", data_only=False)`, change the requested ranges, and save a separate result. Avoid a DataFrame round trip when workbook features must survive.\\n' +\n    '\\n' +\n    'Writing a formula does not calculate its result. openpyxl and XlsxWriter do not evaluate formulas; `data_only=True` returns cached values that may be absent or stale. When calculated results must be refreshed, call `office_recalculate` with `format: \"xlsx\"`, and copy the returned temporary workbook to a new file in the workspace. Open the copy with openpyxl using both `data_only=False` and `data_only=True` to check that formulas survive and cached values match the expected results. If recalculation fails, preserve the original workbook and report that calculated values were not refreshed. Do not replace requested formulas with constants or report stale or absent cached values as newly calculated results.\\n' +\n    '\\n' +\n    'Do not rename `.xls`, `.xlsb`, encrypted files, or macro-enabled files to `.xlsx`. They need an appropriate supported operation. `keep_vba=True` can preserve VBA package content in an XLSM workflow, but does not execute or edit macros. Preserve the original and verify each workbook feature the task depends on.\\n' +\n    '\\n' +\n    '## Check and deliver\\n' +\n    '\\n' +\n    'Reopen the saved workbook with openpyxl and verify the requested sheet names, values, types, formulas, styles, and totals against the source data, including unchanged content that must be preserved. These checks do not calculate formulas or judge workbook appearance.\\n' +\n    '\\n' +\n    'Call `office_render` when the user asks for images or visual inspection, or when the task targets appearance, such as cell fill colors, text alignment, spacing, or sizing; a request to skip rendering takes precedence. Do not render new workbooks or data or formula edits merely to verify accompanying bold headers or number formats; check thesLine truncated
 {"type":"step/end","data":{"turn":1,"step":1}}
 {"type":"step/start","data":{"turn":1,"step":2}}
 {"type":"system/message","data":{"turn":1,"step":2,"message":{"role":"system","content":[{"type":"text","text":"{{system}}"}],"source":{"kind":"system-prompt"},"id":"{{message:8}}"}},"sourceEventSeqs":[7],"surfaceOp":{"op":"replace","startSeq":7,"endSeq":7}}
 {"type":"request/header","data":{"header":{"config":{"provider":"deepseek-official","model":"deepseek-v4-flash"},"tools":"{{tools}}"},"reason":"series"}}
-{"type":"assistant/message","data":{"turn":1,"step":2,"message":{"role":"assistant","content":[{"type":"tool-call","id":"call_office_2","name":"run_code","arguments":"{\"code\":\"inspect(await tools.office_tools({\\\"action\\\":\\\"load\\\"}));\",\"description\":\"Use the Office workflow\"}"}],"source":{"kind":"model","provider":"deepseek-official","model":"deepseek-v4-flash"},"id":"{{message:9}}"},"stream":[{"type":"chunk","time":1794204019263,"chunk":{"type":"block-start","index":0,"blockType":"tool-call"}},{"type":"chunk","time":1794204019263,"chunk":{"type":"block-end","index":0,"block":{"type":"tool-call","id":"call_office_2","name":"run_code","arguments":"{\"code\":\"inspect(await tools.office_tools({\\\"action\\\":\\\"load\\\"}));\",\"description\":\"Use the Office workflow\"}"}}},{"type":"chunk","time":1794204019264,"chunk":{"type":"finish","reason":{"kind":"tool-calls"}}}]},"surfaceOp":"append"}
+{"type":"assistant/message","data":{"turn":1,"step":2,"message":{"role":"assistant","content":[{"type":"tool-call","id":"call_office_2","name":"run_code","arguments":"{\"code\":\"inspect(await tools.office_tools({\\\"action\\\":\\\"load\\\"}));\",\"description\":\"Use the Office workflow\"}"}],"source":{"kind":"model","provider":"deepseek-official","model":"deepseek-v4-flash"},"id":"{{message:9}}"},"stream":[{"type":"chunk","time":1794283521671,"chunk":{"type":"block-start","index":0,"blockType":"tool-call"}},{"type":"chunk","time":1794283521671,"chunk":{"type":"block-end","index":0,"block":{"type":"tool-call","id":"call_office_2","name":"run_code","arguments":"{\"code\":\"inspect(await tools.office_tools({\\\"action\\\":\\\"load\\\"}));\",\"description\":\"Use the Office workflow\"}"}}},{"type":"chunk","time":1794283521671,"chunk":{"type":"finish","reason":{"kind":"tool-calls"}}}]},"surfaceOp":"append"}
 {"type":"tool/call","data":{"turn":1,"step":2,"callId":"call_office_2","name":"run_code","arguments":"{\"code\":\"inspect(await tools.office_tools({\\\"action\\\":\\\"load\\\"}));\",\"description\":\"Use the Office workflow\"}"}}
 {"type":"tool/result","data":{"turn":1,"step":2,"message":{"role":"tool","source":{"kind":"tool","callId":"call_office_2"},"toolCallId":"call_office_2","content":[{"type":"text","text":"Error: code run failed (exception): TypeError: tools.office_tools is not a function\nFile sandbox: danger-full-access."}],"isError":true,"id":"{{message:10}}"},"error":{"name":"CodeRunFailedError","code":"CODE_RUN_FAILED"}},"sourceEventSeqs":[25],"surfaceOp":"append"}
 {"type":"step/end","data":{"turn":1,"step":2}}
 {"type":"step/start","data":{"turn":1,"step":3}}
-{"type":"assistant/message","data":{"turn":1,"step":3,"message":{"role":"assistant","content":[{"type":"tool-call","id":"call_office_3","name":"run_code","arguments":"{\"code\":\"inspect(await tools.render_document({\\\"input\\\":\\\"report.docx\\\",\\\"outputDirectory\\\":\\\"preview\\\",\\\"pages\\\":[1]}));\",\"description\":\"Use the Office workflow\"}"}],"source":{"kind":"model","provider":"deepseek-official","model":"deepseek-v4-flash"},"id":"{{message:11}}"},"stream":[{"type":"chunk","time":1794204019400,"chunk":{"type":"block-start","index":0,"blockType":"tool-call"}},{"type":"chunk","time":1794204019400,"chunk":{"type":"block-end","index":0,"block":{"type":"tool-call","id":"call_office_3","name":"run_code","arguments":"{\"code\":\"inspect(await tools.render_document({\\\"input\\\":\\\"report.docx\\\",\\\"outputDirectory\\\":\\\"preview\\\",\\\"pages\\\":[1]}));\",\"description\":\"Use the Office workflow\"}"}}},{"type":"chunk","time":1794204019400,"chunk":{"type":"finish","reason":{"kind":"tool-calls"}}}]},"surfaceOp":"append"}
+{"type":"assistant/message","data":{"turn":1,"step":3,"message":{"role":"assistant","content":[{"type":"tool-call","id":"call_office_3","name":"run_code","arguments":"{\"code\":\"inspect(await tools.render_document({\\\"input\\\":\\\"report.docx\\\",\\\"outputDirectory\\\":\\\"preview\\\",\\\"pages\\\":[1]}));\",\"description\":\"Use the Office workflow\"}"}],"source":{"kind":"model","provider":"deepseek-official","model":"deepseek-v4-flash"},"id":"{{message:11}}"},"stream":[{"type":"chunk","time":1794283521805,"chunk":{"type":"block-start","index":0,"blockType":"tool-call"}},{"type":"chunk","time":1794283521805,"chunk":{"type":"block-end","index":0,"block":{"type":"tool-call","id":"call_office_3","name":"run_code","arguments":"{\"code\":\"inspect(await tools.render_document({\\\"input\\\":\\\"report.docx\\\",\\\"outputDirectory\\\":\\\"preview\\\",\\\"pages\\\":[1]}));\",\"description\":\"Use the Office workflow\"}"}}},{"type":"chunk","time":1794283521805,"chunk":{"type":"finish","reason":{"kind":"tool-calls"}}}]},"surfaceOp":"append"}
 {"type":"tool/call","data":{"turn":1,"step":3,"callId":"call_office_3","name":"run_code","arguments":"{\"code\":\"inspect(await tools.render_document({\\\"input\\\":\\\"report.docx\\\",\\\"outputDirectory\\\":\\\"preview\\\",\\\"pages\\\":[1]}));\",\"description\":\"Use the Office workflow\"}"}}
 {"type":"tool/result","data":{"turn":1,"step":3,"message":{"role":"tool","source":{"kind":"tool","callId":"call_office_3"},"toolCallId":"call_office_3","content":[{"type":"text","text":"Error: code run failed (exception): TypeError: tools.render_document is not a function\nFile sandbox: danger-full-access."}],"isError":true,"id":"{{message:12}}"},"error":{"name":"CodeRunFailedError","code":"CODE_RUN_FAILED"}},"sourceEventSeqs":[30],"surfaceOp":"append"}
 {"type":"step/end","data":{"turn":1,"step":3}}
 {"type":"step/start","data":{"turn":1,"step":4}}
-{"type":"assistant/message","data":{"turn":1,"step":4,"message":{"role":"assistant","content":[{"type":"tool-call","id":"call_office_4","name":"run_code","arguments":"{\"code\":\"inspect(await tools.office_tools({\\\"action\\\":\\\"remove\\\"}));\",\"description\":\"Use the Office workflow\"}"}],"source":{"kind":"model","provider":"deepseek-official","model":"deepseek-v4-flash"},"id":"{{message:13}}"},"stream":[{"type":"chunk","time":1794204019528,"chunk":{"type":"block-start","index":0,"blockType":"tool-call"}},{"type":"chunk","time":1794204019528,"chunk":{"type":"block-end","index":0,"block":{"type":"tool-call","id":"call_office_4","name":"run_code","arguments":"{\"code\":\"inspect(await tools.office_tools({\\\"action\\\":\\\"remove\\\"}));\",\"description\":\"Use the Office workflow\"}"}}},{"type":"chunk","time":1794204019528,"chunk":{"type":"finish","reason":{"kind":"tool-calls"}}}]},"surfaceOp":"append"}
+{"type":"assistant/message","data":{"turn":1,"step":4,"message":{"role":"assistant","content":[{"type":"tool-call","id":"call_office_4","name":"run_code","arguments":"{\"code\":\"inspect(await tools.office_tools({\\\"action\\\":\\\"remove\\\"}));\",\"description\":\"Use the Office workflow\"}"}],"source":{"kind":"model","provider":"deepseek-official","model":"deepseek-v4-flash"},"id":"{{message:13}}"},"stream":[{"type":"chunk","time":1794283521943,"chunk":{"type":"block-start","index":0,"blockType":"tool-call"}},{"type":"chunk","time":1794283521943,"chunk":{"type":"block-end","index":0,"block":{"type":"tool-call","id":"call_office_4","name":"run_code","arguments":"{\"code\":\"inspect(await tools.office_tools({\\\"action\\\":\\\"remove\\\"}));\",\"description\":\"Use the Office workflow\"}"}}},{"type":"chunk","time":1794283521943,"chunk":{"type":"finish","reason":{"kind":"tool-calls"}}}]},"surfaceOp":"append"}
 {"type":"tool/call","data":{"turn":1,"step":4,"callId":"call_office_4","name":"run_code","arguments":"{\"code\":\"inspect(await tools.office_tools({\\\"action\\\":\\\"remove\\\"}));\",\"description\":\"Use the Office workflow\"}"}}
 {"type":"tool/result","data":{"turn":1,"step":4,"message":{"role":"tool","source":{"kind":"tool","callId":"call_office_4"},"toolCallId":"call_office_4","content":[{"type":"text","text":"Error: code run failed (exception): TypeError: tools.office_tools is not a function\nFile sandbox: danger-full-access."}],"isError":true,"id":"{{message:14}}"},"error":{"name":"CodeRunFailedError","code":"CODE_RUN_FAILED"}},"sourceEventSeqs":[35],"surfaceOp":"append"}
 {"type":"step/end","data":{"turn":1,"step":4}}
 {"type":"step/start","data":{"turn":1,"step":5}}
-{"type":"assistant/message","data":{"turn":1,"step":5,"message":{"role":"assistant","content":[{"type":"text","text":"DONE"}],"source":{"kind":"model","provider":"deepseek-official","model":"deepseek-v4-flash"},"id":"{{message:15}}"},"usage":{"inputTokens":12,"outputTokens":2,"cacheReadTokens":10},"stream":[{"type":"chunk","time":1794204019655,"chunk":{"type":"block-start","index":0,"blockType":"text"}},{"type":"chunk","time":1794204019655,"chunk":{"type":"block-end","index":0,"block":{"type":"text","text":"DONE"}}},{"type":"chunk","time":1794204019655,"chunk":{"type":"usage","usage":{"inputTokens":12,"outputTokens":2,"cacheReadTokens":10}}},{"type":"chunk","time":1794204019655,"chunk":{"type":"finish","reason":{"kind":"stop"}}}]},"surfaceOp":"append"}
+{"type":"assistant/message","data":{"turn":1,"step":5,"message":{"role":"assistant","content":[{"type":"text","text":"DONE"}],"source":{"kind":"model","provider":"deepseek-official","model":"deepseek-v4-flash"},"id":"{{message:15}}"},"usage":{"inputTokens":12,"outputTokens":2,"cacheReadTokens":10},"stream":[{"type":"chunk","time":1794283522075,"chunk":{"type":"block-start","index":0,"blockType":"text"}},{"type":"chunk","time":1794283522075,"chunk":{"type":"block-end","index":0,"block":{"type":"text","text":"DONE"}}},{"type":"chunk","time":1794283522075,"chunk":{"type":"usage","usage":{"inputTokens":12,"outputTokens":2,"cacheReadTokens":10}}},{"type":"chunk","time":1794283522075,"chunk":{"type":"finish","reason":{"kind":"stop"}}}]},"surfaceOp":"append"}
 {"type":"step/end","data":{"turn":1,"step":5}}
 {"type":"turn/end","data":{"turn":1,"reason":{"kind":"completed"}}}

snapshots/session/ptc-console-enabled/session.v4.jsonl

@@ -14,7 +14,7 @@
 {"type":"session/title","data":{"title":"Inspect literal strings and selected","messageSeqs":[8],"source":{"kind":"fallback"}}}
 {"type":"assistant/message","data":{"turn":1,"step":1,"message":{"role":"assistant","content":[{"type":"tool-call","id":"call_00_UiQPVqoELyzBZCY5pm1z7875","name":"run_code","arguments":"{\"code\":\"inspect(\\\"selected\\\", { visible: 42 }); for (const method of [\\\"log\\\", \\\"info\\\", \\\"warn\\\", \\\"error\\\", \\\"debug\\\"]) console[method](method); console.dir({ dependency: true }); return 42n;\",\"description\":\"Inspect selected program values\"}"}],"source":{"kind":"model","provider":"deepseek-official","model":"deepseek-v4-flash"},"id":"{{message:4}}"},"usage":{"inputTokens":10,"outputTokens":5},"stream":[{"type":"chunk","time":1794193200000,"chunk":{"type":"block-start","index":0,"blockType":"tool-call"}},{"type":"chunk","time":1794193200001,"chunk":{"type":"block-end","index":0,"block":{"type":"tool-call","id":"call_00_UiQPVqoELyzBZCY5pm1z7875","name":"run_code","arguments":"{\"code\":\"inspect(\\\"selected\\\", { visible: 42 }); for (const method of [\\\"log\\\", \\\"info\\\", \\\"warn\\\", \\\"error\\\", \\\"debug\\\"]) console[method](method); console.dir({ dependency: true }); return 42n;\",\"description\":\"Inspect selected program values\"}"}}},{"type":"chunk","time":1794193200002,"chunk":{"type":"usage","usage":{"inputTokens":10,"outputTokens":5}}},{"type":"chunk","time":1794193200003,"chunk":{"type":"finish","reason":{"kind":"tool-calls"}}}]},"surfaceOp":"append"}
 {"type":"tool/call","data":{"turn":1,"step":1,"callId":"call_00_UiQPVqoELyzBZCY5pm1z7875","name":"run_code","arguments":"{\"code\":\"inspect(\\\"selected\\\", { visible: 42 }); for (const method of [\\\"log\\\", \\\"info\\\", \\\"warn\\\", \\\"error\\\", \\\"debug\\\"]) console[method](method); console.dir({ dependency: true }); return 42n;\",\"description\":\"Inspect selected program values\"}"}}
-{"type":"tool/result","data":{"turn":1,"step":1,"message":{"role":"tool","source":{"kind":"tool","callId":"call_00_UiQPVqoELyzBZCY5pm1z7875"},"toolCallId":"call_00_UiQPVqoELyzBZCY5pm1z7875","content":[{"type":"text","text":"selected {\n  \"visible\": 42\n}\nlog\n\ninfo\n\nwarn\n\nerror\n\ndebug\n\n{ dependency: true }\n"}],"isError":false,"id":"{{message:5}}"}},"sourceEventSeqs":[14],"surfaceOp":"append"}
+{"type":"tool/result","data":{"turn":1,"step":1,"message":{"role":"tool","source":{"kind":"tool","callId":"call_00_UiQPVqoELyzBZCY5pm1z7875"},"toolCallId":"call_00_UiQPVqoELyzBZCY5pm1z7875","content":[{"type":"text","text":"selected { visible: 42 }\nlog\n\ninfo\n\nwarn\n\nerror\n\ndebug\n\n{ dependency: true }\n"}],"isError":false,"id":"{{message:5}}"}},"sourceEventSeqs":[14],"surfaceOp":"append"}
 {"type":"step/end","data":{"turn":1,"step":1}}
 {"type":"step/start","data":{"turn":1,"step":2}}
 {"type":"assistant/message","data":{"turn":1,"step":2,"message":{"role":"assistant","content":[{"type":"text","text":"INSPECT_DONE"}],"source":{"kind":"model","provider":"deepseek-official","model":"deepseek-v4-flash"},"id":"{{message:6}}"},"usage":{"inputTokens":10,"outputTokens":5},"stream":[{"type":"chunk","time":1794193200000,"chunk":{"type":"block-start","index":0,"blockType":"text"}},{"type":"chunk","time":1794193200001,"chunk":{"type":"block-end","index":0,"block":{"type":"text","text":"INSPECT_DONE"}}},{"type":"chunk","time":1794193200002,"chunk":{"type":"usage","usage":{"inputTokens":10,"outputTokens":5}}},{"type":"chunk","time":1794193200003,"chunk":{"type":"finish","reason":{"kind":"stop"}}}]},"surfaceOp":"append"}

snapshots/session/ptc-inspect-output/session.v4.jsonl

@@ -12,9 +12,9 @@
 {"type":"request/header","data":{"header":{"config":{"provider":"deepseek-official","model":"deepseek-v4-flash"},"tools":"{{tools}}"},"reason":"initial"}}
 {"type":"request/context","data":{"provider":"deepseek-official","model":"deepseek-v4-flash"}}
 {"type":"session/title","data":{"title":"Inspect literal strings and selected","messageSeqs":[8],"source":{"kind":"fallback"}}}
-{"type":"assistant/message","data":{"turn":1,"step":1,"message":{"role":"assistant","content":[{"type":"tool-call","id":"call_00_UiQPVqoELyzBZCY5pm1z7875","name":"run_code","arguments":"{\"code\":\"inspect({ rows: Array.from({ length: 101 }, (_, index) => index) }); try { inspect(new Uint8Array([1])); } catch (error) { inspect(error.message); } inspect(\\\"%s\\\", \\\"literal\\\", { visible: 42 }); { const console = { value: 17 }; inspect(console.value) }; return { get hidden() { throw new Error(\\\"completion getter ran\\\") } };\",\"description\":\"Inspect selected program values\"}"}],"source":{"kind":"model","provider":"deepseek-official","model":"deepseek-v4-flash"},"id":"{{message:4}}"},"usage":{"inputTokens":10,"outputTokens":5},"stream":[{"type":"chunk","time":1794193200000,"chunk":{"type":"block-start","index":0,"blockType":"tool-call"}},{"type":"chunk","time":1794193200001,"chunk":{"type":"block-end","index":0,"block":{"type":"tool-call","id":"call_00_UiQPVqoELyzBZCY5pm1z7875","name":"run_code","arguments":"{\"code\":\"inspect({ rows: Array.from({ length: 101 }, (_, index) => index) }); try { inspect(new Uint8Array([1])); } catch (error) { inspect(error.message); } inspect(\\\"%s\\\", \\\"literal\\\", { visible: 42 }); { const console = { value: 17 }; inspect(console.value) }; return { get hidden() { throw new Error(\\\"completion getter ran\\\") } };\",\"description\":\"Inspect selected program values\"}"}}},{"type":"chunk","time":1794193200002,"chunk":{"type":"usage","usage":{"inputTokens":10,"outputTokens":5}}},{"type":"chunk","time":1794193200003,"chunk":{"type":"finish","reason":{"kind":"tool-calls"}}}]},"surfaceOp":"append"}
-{"type":"tool/call","data":{"turn":1,"step":1,"callId":"call_00_UiQPVqoELyzBZCY5pm1z7875","name":"run_code","arguments":"{\"code\":\"inspect({ rows: Array.from({ length: 101 }, (_, index) => index) }); try { inspect(new Uint8Array([1])); } catch (error) { inspect(error.message); } inspect(\\\"%s\\\", \\\"literal\\\", { visible: 42 }); { const console = { value: 17 }; inspect(console.value) }; return { get hidden() { throw new Error(\\\"completion getter ran\\\") } };\",\"description\":\"Inspect selected program values\"}"}}
-{"type":"tool/result","data":{"turn":1,"step":1,"message":{"role":"tool","source":{"kind":"tool","callId":"call_00_UiQPVqoELyzBZCY5pm1z7875"},"toolCallId":"call_00_UiQPVqoELyzBZCY5pm1z7875","content":[{"type":"text","text":"{\n  \"rows\": [\n    0,\n    1,\n    2,\n    3,\n    4,\n    5,\n    6,\n    7,\n    8,\n    9,\n    10,\n    11,\n    12,\n    13,\n    14,\n    15,\n    16,\n    17,\n    18,\n    19,\n    20,\n    21,\n    22,\n    23,\n    24,\n    25,\n    26,\n    27,\n    28,\n    29,\n    30,\n    31,\n    32,\n    33,\n    34,\n    35,\n    36,\n    37,\n    38,\n    39,\n    40,\n    41,\n    42,\n    43,\n    44,\n    45,\n    46,\n    47,\n    48,\n    49,\n    50,\n    51,\n    52,\n    53,\n    54,\n    55,\n    56,\n    57,\n    58,\n    59,\n    60,\n    61,\n    62,\n    63,\n    64,\n    65,\n    66,\n    67,\n    68,\n    69,\n    70,\n    71,\n    72,\n    73,\n    74,\n    75,\n    76,\n    77,\n    78,\n    79,\n    80,\n    81,\n    82,\n    83,\n    84,\n    85,\n    86,\n    87,\n    88,\n    89,\n    90,\n    91,\n    92,\n    93,\n    94,\n    95,\n    96,\n    97,\n    98,\n    99,\n    100\n  ]\n}\ninspect argument 1 must be lossless JSON or an Image; omit undefined properties and pass error.message for Errors\n%s literal {\n  \"visible\": 42\n}\n17"}],"isError":false,"id":"{{message:5}}"}},"sourceEventSeqs":[14],"surfaceOp":"append"}
+{"type":"assistant/message","data":{"turn":1,"step":1,"message":{"role":"assistant","content":[{"type":"tool-call","id":"call_00_UiQPVqoELyzBZCY5pm1z7875","name":"run_code","arguments":"{\"code\":\"inspect({ rows: Array.from({ length: 101 }, (_, index) => index) }); inspect({ path: \\\"/file.ts\\\", operation: undefined }); inspect(new Uint8Array([1])); inspect(\\\"%s\\\", \\\"literal\\\", { visible: 42 }); { const console = { value: 17 }; inspect(console.value) }; return { get hidden() { throw new Error(\\\"completion getter ran\\\") } };\",\"description\":\"Inspect selected program values\"}"}],"source":{"kind":"model","provider":"deepseek-official","model":"deepseek-v4-flash"},"id":"{{message:4}}"},"usage":{"inputTokens":10,"outputTokens":5},"stream":[{"type":"chunk","time":1794193200000,"chunk":{"type":"block-start","index":0,"blockType":"tool-call"}},{"type":"chunk","time":1794193200001,"chunk":{"type":"block-end","index":0,"block":{"type":"tool-call","id":"call_00_UiQPVqoELyzBZCY5pm1z7875","name":"run_code","arguments":"{\"code\":\"inspect({ rows: Array.from({ length: 101 }, (_, index) => index) }); inspect({ path: \\\"/file.ts\\\", operation: undefined }); inspect(new Uint8Array([1])); inspect(\\\"%s\\\", \\\"literal\\\", { visible: 42 }); { const console = { value: 17 }; inspect(console.value) }; return { get hidden() { throw new Error(\\\"completion getter ran\\\") } };\",\"description\":\"Inspect selected program values\"}"}}},{"type":"chunk","time":1794193200002,"chunk":{"type":"usage","usage":{"inputTokens":10,"outputTokens":5}}},{"type":"chunk","time":1794193200003,"chunk":{"type":"finish","reason":{"kind":"tool-calls"}}}]},"surfaceOp":"append"}
+{"type":"tool/call","data":{"turn":1,"step":1,"callId":"call_00_UiQPVqoELyzBZCY5pm1z7875","name":"run_code","arguments":"{\"code\":\"inspect({ rows: Array.from({ length: 101 }, (_, index) => index) }); inspect({ path: \\\"/file.ts\\\", operation: undefined }); inspect(new Uint8Array([1])); inspect(\\\"%s\\\", \\\"literal\\\", { visible: 42 }); { const console = { value: 17 }; inspect(console.value) }; return { get hidden() { throw new Error(\\\"completion getter ran\\\") } };\",\"description\":\"Inspect selected program values\"}"}}
+{"type":"tool/result","data":{"turn":1,"step":1,"message":{"role":"tool","source":{"kind":"tool","callId":"call_00_UiQPVqoELyzBZCY5pm1z7875"},"toolCallId":"call_00_UiQPVqoELyzBZCY5pm1z7875","content":[{"type":"text","text":"{\n  rows: [\n     0,  1,  2,  3,   4,  5,  6,  7,  8,  9, 10, 11,\n    12, 13, 14, 15,  16, 17, 18, 19, 20, 21, 22, 23,\n    24, 25, 26, 27,  28, 29, 30, 31, 32, 33, 34, 35,\n    36, 37, 38, 39,  40, 41, 42, 43, 44, 45, 46, 47,\n    48, 49, 50, 51,  52, 53, 54, 55, 56, 57, 58, 59,\n    60, 61, 62, 63,  64, 65, 66, 67, 68, 69, 70, 71,\n    72, 73, 74, 75,  76, 77, 78, 79, 80, 81, 82, 83,\n    84, 85, 86, 87,  88, 89, 90, 91, 92, 93, 94, 95,\n    96, 97, 98, 99, 100\n  ]\n}\n{ path: '/file.ts', operation: undefined }\nUint8Array(1) [ 1 ]\n%s literal { visible: 42 }\n17"}],"isError":false,"id":"{{message:5}}"}},"sourceEventSeqs":[14],"surfaceOp":"append"}
 {"type":"step/end","data":{"turn":1,"step":1}}
 {"type":"step/start","data":{"turn":1,"step":2}}
 {"type":"assistant/message","data":{"turn":1,"step":2,"message":{"role":"assistant","content":[{"type":"text","text":"INSPECT_DONE"}],"source":{"kind":"model","provider":"deepseek-official","model":"deepseek-v4-flash"},"id":"{{message:6}}"},"usage":{"inputTokens":10,"outputTokens":5},"stream":[{"type":"chunk","time":1794193200000,"chunk":{"type":"block-start","index":0,"blockType":"text"}},{"type":"chunk","time":1794193200001,"chunk":{"type":"block-end","index":0,"block":{"type":"text","text":"INSPECT_DONE"}}},{"type":"chunk","time":1794193200002,"chunk":{"type":"usage","usage":{"inputTokens":10,"outputTokens":5}}},{"type":"chunk","time":1794193200003,"chunk":{"type":"finish","reason":{"kind":"stop"}}}]},"surfaceOp":"append"}

snapshots/session/ptc-node-read-only/session.v4.jsonl

@@ -18,7 +18,7 @@
 {"type":"job/registered","data":{"jobId":"{{id:1}}","owner":"{{session:1}}","kind":"bash","label":"pwd","startedAt":1791427867102,"toolCall":{"callId":"call_node_policy:ptc:1","rootCallId":"call_node_policy"}}}
 {"type":"job/settled","data":{"jobId":"{{id:1}}","owner":"{{session:1}}","kind":"bash","label":"pwd","startedAt":1791427867102,"toolCall":{"callId":"call_node_policy:ptc:1","rootCallId":"call_node_policy"},"finishedAt":1791427867112,"status":"completed","cause":"producer","producerReported":true,"detail":"exit code: 0"}}
 {"type":"tool/ptc-dispatch","data":{"rootCallId":"call_node_policy","parentCallId":"call_node_policy","subCallId":"call_node_policy:ptc:1","name":"bash","arguments":{"command":"pwd","description":"Report the current workspace directory"},"isError":false,"content":[{"type":"text","text":"{{cwd}}\n"}],"meta":{"cwd":"{{cwd}}"}}}
-{"type":"tool/result","data":{"turn":1,"step":1,"message":{"role":"tool","source":{"kind":"tool","callId":"call_node_policy"},"toolCallId":"call_node_policy","content":[{"type":"text","text":"{\n  \"denied\": true,\n  \"emptyEnvironment\": true,\n  \"sameWorkspace\": true\n}"}],"isError":false,"id":"{{message:5}}"}},"sourceEventSeqs":[14],"surfaceOp":"append"}
+{"type":"tool/result","data":{"turn":1,"step":1,"message":{"role":"tool","source":{"kind":"tool","callId":"call_node_policy"},"toolCallId":"call_node_policy","content":[{"type":"text","text":"{ denied: true, emptyEnvironment: true, sameWorkspace: true }"}],"isError":false,"id":"{{message:5}}"}},"sourceEventSeqs":[14],"surfaceOp":"append"}
 {"type":"step/end","data":{"turn":1,"step":1}}
 {"type":"step/start","data":{"turn":1,"step":2}}
 {"type":"assistant/message","data":{"turn":1,"step":2,"message":{"role":"assistant","content":[{"type":"text","text":"Node policy and execution context verified."}],"source":{"kind":"model","provider":"deepseek-official","model":"deepseek-v4-flash"},"id":"{{message:6}}"},"usage":{"inputTokens":10,"outputTokens":2},"stream":[{"type":"chunk","time":1791427867146,"chunk":{"type":"block-start","index":0,"blockType":"text"}},{"type":"text-chunks","time0":1791427867146,"index":0,"dt":[],"texts":["Node policy and execution context verified."]},{"type":"chunk","time":1791427867146,"chunk":{"type":"block-end","index":0,"block":{"type":"text","text":"Node policy and execution context verified."}}},{"type":"chunk","time":1791427867146,"chunk":{"type":"usage","usage":{"inputTokens":10,"outputTokens":2}}},{"type":"chunk","time":1791427867146,"chunk":{"type":"finish","reason":{"kind":"stop"}}}]},"surfaceOp":"append"}

snapshots/session/ptc-node-workspace/session.v4.jsonl

@@ -18,7 +18,7 @@
 {"type":"job/registered","data":{"jobId":"{{id:1}}","owner":"{{session:1}}","kind":"bash","label":"pwd","startedAt":1791427868009,"toolCall":{"callId":"call_node_policy:ptc:1","rootCallId":"call_node_policy"}}}
 {"type":"job/settled","data":{"jobId":"{{id:1}}","owner":"{{session:1}}","kind":"bash","label":"pwd","startedAt":1791427868009,"toolCall":{"callId":"call_node_policy:ptc:1","rootCallId":"call_node_policy"},"finishedAt":1791427868019,"status":"completed","cause":"producer","producerReported":true,"detail":"exit code: 0"}}
 {"type":"tool/ptc-dispatch","data":{"rootCallId":"call_node_policy","parentCallId":"call_node_policy","subCallId":"call_node_policy:ptc:1","name":"bash","arguments":{"command":"pwd","description":"Report the current workspace directory"},"isError":false,"content":[{"type":"text","text":"{{cwd}}\n"}],"meta":{"cwd":"{{cwd}}"}}}
-{"type":"tool/result","data":{"turn":1,"step":1,"message":{"role":"tool","source":{"kind":"tool","callId":"call_node_policy"},"toolCallId":"call_node_policy","content":[{"type":"text","text":"{\n  \"written\": true,\n  \"emptyEnvironment\": true,\n  \"sameWorkspace\": true\n}"}],"isError":false,"id":"{{message:5}}"}},"sourceEventSeqs":[14],"surfaceOp":"append"}
+{"type":"tool/result","data":{"turn":1,"step":1,"message":{"role":"tool","source":{"kind":"tool","callId":"call_node_policy"},"toolCallId":"call_node_policy","content":[{"type":"text","text":"{ written: true, emptyEnvironment: true, sameWorkspace: true }"}],"isError":false,"id":"{{message:5}}"}},"sourceEventSeqs":[14],"surfaceOp":"append"}
 {"type":"step/end","data":{"turn":1,"step":1}}
 {"type":"step/start","data":{"turn":1,"step":2}}
 {"type":"assistant/message","data":{"turn":1,"step":2,"message":{"role":"assistant","content":[{"type":"text","text":"Node policy and execution context verified."}],"source":{"kind":"model","provider":"deepseek-official","model":"deepseek-v4-flash"},"id":"{{message:6}}"},"usage":{"inputTokens":10,"outputTokens":2},"stream":[{"type":"chunk","time":1791427868053,"chunk":{"type":"block-start","index":0,"blockType":"text"}},{"type":"text-chunks","time0":1791427868053,"index":0,"dt":[],"texts":["Node policy and execution context verified."]},{"type":"chunk","time":1791427868053,"chunk":{"type":"block-end","index":0,"block":{"type":"text","text":"Node policy and execution context verified."}}},{"type":"chunk","time":1791427868053,"chunk":{"type":"usage","usage":{"inputTokens":10,"outputTokens":2}}},{"type":"chunk","time":1791427868053,"chunk":{"type":"finish","reason":{"kind":"stop"}}}]},"surfaceOp":"append"}

snapshots/session/ptc-persistent/system-prompt.expected.md

@@ -46,7 +46,7 @@ Inside the program:
 - Call tools as `await tools.name(args)` — quoted access for exotic names: `tools["my-tool"](args)`. Every call resolves to its declared value, which may be an Image. Tool arguments must be lossless JSON.
 - A FAILED tool call rejects with `ToolCallError`, whose `toolName` identifies the failed tool and whose `message` is human-readable — `try/catch` it to handle and continue.
 - Independent read-only calls MAY overlap under `Promise.all` (safe calls run concurrently; mutating calls run alone, in submission order). Sequence dependent work with `await`.
-- Show selected results with `inspect(...values)`. Each call accepts JSON values and Images, including Images nested in arrays and objects. Images expose a readable file path and its pixel dimensions. Extract just what you need.
+- Show selected values and images with `inspect(...values)`; every other intermediate result stays out of the conversation, so extract just what you need.
 
 Program-only SDK bindings:
 

snapshots/session/ptc-policy-change/system-prompt.expected.md

@@ -46,7 +46,7 @@ Inside the program:
 - Call tools as `await tools.name(args)` — quoted access for exotic names: `tools["my-tool"](args)`. Every call resolves to its declared value, which may be an Image. Tool arguments must be lossless JSON.
 - A FAILED tool call rejects with `ToolCallError`, whose `toolName` identifies the failed tool and whose `message` is human-readable — `try/catch` it to handle and continue.
 - Independent read-only calls MAY overlap under `Promise.all` (safe calls run concurrently; mutating calls run alone, in submission order). Sequence dependent work with `await`.
-- Show selected results with `inspect(...values)`. Each call accepts JSON values and Images, including Images nested in arrays and objects. Images expose a readable file path and its pixel dimensions. Extract just what you need.
+- Show selected values and images with `inspect(...values)`; every other intermediate result stays out of the conversation, so extract just what you need.
 
 Program-only SDK bindings:
 

snapshots/session/ptc-read-image/session.v4.jsonl

@@ -22,7 +22,7 @@
 {"type":"tool/ptc-dispatch","data":{"rootCallId":"code-image-call","parentCallId":"code-image-call","subCallId":"code-image-call:ptc:2","name":"read_image","arguments":{"file_path":"red.png"},"isError":false,"content":[{"type":"text","text":"<path>{{cwd}}/red.png</path>\n<type>image</type>\n<content>\nimage/png image, 1x1 px, 69 bytes\n</content>"},{"type":"image","attachment":{"attachmentId":"sha256:b1ff9c8ea3a780bad09b346c423d2d0e46815926879b18e841d928376a946640","mediaType":"image/png","bytes":69,"width":1,"height":1,"name":"red.png"}}],"meta":{"path":"{{cwd}}/red.png"}}}
 {"type":"tool/ptc-dispatch-start","data":{"rootCallId":"code-image-call","parentCallId":"code-image-call","subCallId":"code-image-call:ptc:3","name":"read_image","arguments":{"file_path":"red.png"}}}
 {"type":"tool/ptc-dispatch","data":{"rootCallId":"code-image-call","parentCallId":"code-image-call","subCallId":"code-image-call:ptc:3","name":"read_image","arguments":{"file_path":"red.png"},"isError":false,"content":[{"type":"text","text":"<path>{{cwd}}/red.png</path>\n<type>image</type>\n<content>\nimage/png image, 1x1 px, 69 bytes\n</content>"},{"type":"image","attachment":{"attachmentId":"sha256:b1ff9c8ea3a780bad09b346c423d2d0e46815926879b18e841d928376a946640","mediaType":"image/png","bytes":69,"width":1,"height":1,"name":"red.png"}}],"meta":{"path":"{{cwd}}/red.png"}}}
-{"type":"tool/result","data":{"turn":1,"step":1,"message":{"role":"tool","source":{"kind":"tool","callId":"code-image-call"},"toolCallId":"code-image-call","content":[{"type":"text","text":"selected {\n  \"attachmentId\": \"sha256:b1ff9c8ea3a780bad09b346c423d2d0e46815926879b18e841d928376a946640\",\n  \"width\": 1,\n  \"height\": 1\n}"},{"type":"image","attachment":{"attachmentId":"sha256:b1ff9c8ea3a780bad09b346c423d2d0e46815926879b18e841d928376a946640","mediaType":"image/png","bytes":69,"width":1,"height":1,"name":"red.png"}}],"isError":false,"id":"{{message:5}}"}},"sourceEventSeqs":[14],"surfaceOp":"append"}
+{"type":"tool/result","data":{"turn":1,"step":1,"message":{"role":"tool","source":{"kind":"tool","callId":"code-image-call"},"toolCallId":"code-image-call","content":[{"type":"text","text":"selected {\n  attachmentId: 'sha256:b1ff9c8ea3a780bad09b346c423d2d0e46815926879b18e841d928376a946640',\n  width: 1,\n  height: 1\n}"},{"type":"image","attachment":{"attachmentId":"sha256:b1ff9c8ea3a780bad09b346c423d2d0e46815926879b18e841d928376a946640","mediaType":"image/png","bytes":69,"width":1,"height":1,"name":"red.png"}}],"isError":false,"id":"{{message:5}}"}},"sourceEventSeqs":[14],"surfaceOp":"append"}
 {"type":"step/end","data":{"turn":1,"step":1}}
 {"type":"step/start","data":{"turn":1,"step":2}}
 {"type":"assistant/message","data":{"turn":1,"step":2,"message":{"role":"assistant","content":[{"type":"text","text":"DONE"}],"source":{"kind":"model","provider":"deepseek-official","model":"deepseek-v4-flash-vision-exp"},"id":"{{message:6}}"},"usage":{"inputTokens":3,"outputTokens":3},"stream":[{"type":"chunk","time":1794204032614,"chunk":{"type":"block-start","index":0,"blockType":"text"}},{"type":"chunk","time":1794204032614,"chunk":{"type":"block-end","index":0,"block":{"type":"text","text":"DONE"}}},{"type":"chunk","time":1794204032614,"chunk":{"type":"usage","usage":{"inputTokens":3,"outputTokens":3}}},{"type":"chunk","time":1794204032614,"chunk":{"type":"finish","reason":{"kind":"stop"}}}]},"surfaceOp":"append"}

snapshots/session/ptc-read-image/system-prompt.expected.md

@@ -46,7 +46,7 @@ Inside the program:
 - Call tools as `await tools.name(args)` — quoted access for exotic names: `tools["my-tool"](args)`. Every call resolves to its declared value, which may be an Image. Tool arguments must be lossless JSON.
 - A FAILED tool call rejects with `ToolCallError`, whose `toolName` identifies the failed tool and whose `message` is human-readable — `try/catch` it to handle and continue.
 - Independent read-only calls MAY overlap under `Promise.all` (safe calls run concurrently; mutating calls run alone, in submission order). Sequence dependent work with `await`.
-- Show selected results with `inspect(...values)`. Each call accepts JSON values and Images, including Images nested in arrays and objects. Images expose a readable file path and its pixel dimensions. Extract just what you need.
+- Show selected values and images with `inspect(...values)`; every other intermediate result stays out of the conversation, so extract just what you need.
 
 Program-only SDK bindings:
 

snapshots/session/ptc-script-imports/session.v4.jsonl

@@ -14,12 +14,12 @@
 {"type":"session/title","data":{"title":"Use Script run_code to import","messageSeqs":[8],"source":{"kind":"fallback"}}}
 {"type":"assistant/message","data":{"turn":1,"step":1,"message":{"role":"assistant","content":[{"type":"tool-call","id":"esm_static","name":"run_code","arguments":"{\"code\":\"var dependency = await import(\\\"./entry.mjs\\\"); inspect(dependency.first())\",\"description\":\"Import ESM and its runner object\"}"}],"source":{"kind":"model","provider":"deepseek-official","model":"deepseek-v4-flash"},"id":"{{message:4}}"},"usage":{"inputTokens":10,"outputTokens":5},"stream":[{"type":"chunk","time":1792561944612,"chunk":{"type":"block-start","index":0,"blockType":"tool-call"}},{"type":"chunk","time":1792561944612,"chunk":{"type":"block-end","index":0,"block":{"type":"tool-call","id":"esm_static","name":"run_code","arguments":"{\"code\":\"var dependency = await import(\\\"./entry.mjs\\\"); inspect(dependency.first())\",\"description\":\"Import ESM and its runner object\"}"}}},{"type":"chunk","time":1792561944612,"chunk":{"type":"usage","usage":{"inputTokens":10,"outputTokens":5}}},{"type":"chunk","time":1792561944612,"chunk":{"type":"finish","reason":{"kind":"tool-calls"}}}]},"surfaceOp":"append"}
 {"type":"tool/call","data":{"turn":1,"step":1,"callId":"esm_static","name":"run_code","arguments":"{\"code\":\"var dependency = await import(\\\"./entry.mjs\\\"); inspect(dependency.first())\",\"description\":\"Import ESM and its runner object\"}"}}
-{"type":"tool/result","data":{"turn":1,"step":1,"message":{"role":"tool","source":{"kind":"tool","callId":"esm_static"},"toolCallId":"esm_static","content":[{"type":"text","text":"[\n  1,\n  true\n]"}],"isError":false,"id":"{{message:5}}"}},"sourceEventSeqs":[14],"surfaceOp":"append"}
+{"type":"tool/result","data":{"turn":1,"step":1,"message":{"role":"tool","source":{"kind":"tool","callId":"esm_static"},"toolCallId":"esm_static","content":[{"type":"text","text":"[ 1, true ]"}],"isError":false,"id":"{{message:5}}"}},"sourceEventSeqs":[14],"surfaceOp":"append"}
 {"type":"step/end","data":{"turn":1,"step":1}}
 {"type":"step/start","data":{"turn":1,"step":2}}
 {"type":"assistant/message","data":{"turn":1,"step":2,"message":{"role":"assistant","content":[{"type":"tool-call","id":"esm_dynamic","name":"run_code","arguments":"{\"code\":\"inspect(await dependency.later())\",\"description\":\"Invoke a retained function with a dynamic import\"}"}],"source":{"kind":"model","provider":"deepseek-official","model":"deepseek-v4-flash"},"id":"{{message:6}}"},"usage":{"inputTokens":10,"outputTokens":5},"stream":[{"type":"chunk","time":1792561944793,"chunk":{"type":"block-start","index":0,"blockType":"tool-call"}},{"type":"chunk","time":1792561944793,"chunk":{"type":"block-end","index":0,"block":{"type":"tool-call","id":"esm_dynamic","name":"run_code","arguments":"{\"code\":\"inspect(await dependency.later())\",\"description\":\"Invoke a retained function with a dynamic import\"}"}}},{"type":"chunk","time":1792561944793,"chunk":{"type":"usage","usage":{"inputTokens":10,"outputTokens":5}}},{"type":"chunk","time":1792561944793,"chunk":{"type":"finish","reason":{"kind":"tool-calls"}}}]},"surfaceOp":"append"}
 {"type":"tool/call","data":{"turn":1,"step":2,"callId":"esm_dynamic","name":"run_code","arguments":"{\"code\":\"inspect(await dependency.later())\",\"description\":\"Invoke a retained function with a dynamic import\"}"}}
-{"type":"tool/result","data":{"turn":1,"step":2,"message":{"role":"tool","source":{"kind":"tool","callId":"esm_dynamic"},"toolCallId":"esm_dynamic","content":[{"type":"text","text":"[\n  true,\n  2\n]"}],"isError":false,"id":"{{message:7}}"}},"sourceEventSeqs":[19],"surfaceOp":"append"}
+{"type":"tool/result","data":{"turn":1,"step":2,"message":{"role":"tool","source":{"kind":"tool","callId":"esm_dynamic"},"toolCallId":"esm_dynamic","content":[{"type":"text","text":"[ true, 2 ]"}],"isError":false,"id":"{{message:7}}"}},"sourceEventSeqs":[19],"surfaceOp":"append"}
 {"type":"step/end","data":{"turn":1,"step":2}}
 {"type":"step/start","data":{"turn":1,"step":3}}
 {"type":"assistant/message","data":{"turn":1,"step":3,"message":{"role":"assistant","content":[{"type":"tool-call","id":"esm_denied","name":"run_code","arguments":"{\"code\":\"inspect(await dependency.denied().catch(error => error.message))\",\"description\":\"Catch a rejected dependency import\"}"}],"source":{"kind":"model","provider":"deepseek-official","model":"deepseek-v4-flash"},"id":"{{message:8}}"},"usage":{"inputTokens":10,"outputTokens":5},"stream":[{"type":"chunk","time":1792561944811,"chunk":{"type":"block-start","index":0,"blockType":"tool-call"}},{"type":"chunk","time":1792561944811,"chunk":{"type":"block-end","index":0,"block":{"type":"tool-call","id":"esm_denied","name":"run_code","arguments":"{\"code\":\"inspect(await dependency.denied().catch(error => error.message))\",\"description\":\"Catch a rejected dependency import\"}"}}},{"type":"chunk","time":1792561944811,"chunk":{"type":"usage","usage":{"inputTokens":10,"outputTokens":5}}},{"type":"chunk","time":1792561944811,"chunk":{"type":"finish","reason":{"kind":"tool-calls"}}}]},"surfaceOp":"append"}

snapshots/session/ptc-script-retained/system-prompt.expected.md

@@ -107,7 +107,7 @@ Inside the program:
 - Call tools as `await tools.name(args)` — quoted access for exotic names: `tools["my-tool"](args)`. Every call resolves to its declared value, which may be an Image. Tool arguments must be lossless JSON.
 - A FAILED tool call rejects with `ToolCallError`, whose `toolName` identifies the failed tool and whose `message` is human-readable — `try/catch` it to handle and continue.
 - Independent read-only calls MAY overlap under `Promise.all` (safe calls run concurrently; mutating calls run alone, in submission order). Sequence dependent work with `await`.
-- Show selected results with `inspect(...values)`. Each call accepts JSON values and Images, including Images nested in arrays and objects. Images expose a readable file path and its pixel dimensions. Extract just what you need.
+- Show selected values and images with `inspect(...values)`; every other intermediate result stays out of the conversation, so extract just what you need.
 
 Program-only SDK bindings:
 

snapshots/session/ptc-turn/system-prompt.expected.md

@@ -46,7 +46,7 @@ Inside the program:
 - Call tools as `await tools.name(args)` — quoted access for exotic names: `tools["my-tool"](args)`. Every call resolves to its declared value, which may be an Image. Tool arguments must be lossless JSON.
 - A FAILED tool call rejects with `ToolCallError`, whose `toolName` identifies the failed tool and whose `message` is human-readable — `try/catch` it to handle and continue.
 - Independent read-only calls MAY overlap under `Promise.all` (safe calls run concurrently; mutating calls run alone, in submission order). Sequence dependent work with `await`.
-- Show selected results with `inspect(...values)`. Each call accepts JSON values and Images, including Images nested in arrays and objects. Images expose a readable file path and its pixel dimensions. Extract just what you need.
+- Show selected values and images with `inspect(...values)`; every other intermediate result stays out of the conversation, so extract just what you need.
 
 Program-only SDK bindings:
 

snapshots/session/ptc-workspace-context/session.v4.jsonl

@@ -17,7 +17,7 @@
 {"type":"tool/call","data":{"turn":1,"step":1,"callId":"call_workspace_read","name":"run_code","arguments":"{\"code\":\"inspect(await tools.read({ file_path: 'nested/task.txt' })); return;\",\"description\":\"Read nested/task.txt\"}"}}
 {"type":"tool/ptc-dispatch-start","data":{"rootCallId":"call_workspace_read","parentCallId":"call_workspace_read","subCallId":"call_workspace_read:ptc:1","name":"read","arguments":{"file_path":"nested/task.txt"}}}
 {"type":"tool/ptc-dispatch","data":{"rootCallId":"call_workspace_read","parentCallId":"call_workspace_read","subCallId":"call_workspace_read:ptc:1","name":"read","arguments":{"file_path":"nested/task.txt"},"isError":false,"content":[{"type":"text","text":"<path>{{cwd}}/nested/task.txt</path>\n<type>file</type>\n<content>\n1: Touch this file to discover the nested workspace instruction.\n\n(End of file - total 1 lines)\n</content>"}],"meta":{"path":"{{cwd}}/nested/task.txt","offset":1,"lines":[{"number":1,"text":"Touch this file to discover the nested workspace instruction."}],"totalLines":1}}}
-{"type":"tool/result","data":{"turn":1,"step":1,"message":{"role":"tool","source":{"kind":"tool","callId":"call_workspace_read"},"toolCallId":"call_workspace_read","content":[{"type":"text","text":"{\n  \"path\": \"{{cwd}}/nested/task.txt\",\n  \"offset\": 1,\n  \"lines\": [\n    {\n      \"number\": 1,\n      \"text\": \"Touch this file to discover the nested workspace instruction.\"\n    }\n  ],\n  \"totalLines\": 1\n}"}],"isError":false,"id":"{{message:6}}"}},"sourceEventSeqs":[15],"surfaceOp":"append"}
+{"type":"tool/result","data":{"turn":1,"step":1,"message":{"role":"tool","source":{"kind":"tool","callId":"call_workspace_read"},"toolCallId":"call_workspace_read","content":[{"type":"text","text":"{\n  path: '{{cwd}}/nested/task.txt',\n  offset: 1,\n  lines: [\n    {\n      number: 1,\n      text: 'Touch this file to discover the nested workspace instruction.'\n    }\n  ],\n  totalLines: 1\n}"}],"isError":false,"id":"{{message:6}}"}},"sourceEventSeqs":[15],"surfaceOp":"append"}
 {"type":"step/end","data":{"turn":1,"step":1}}
 {"type":"agent/inbox/spliced","data":{"target":"next-step","start":0,"inserted":[{"content":[{"type":"text","text":"<system-reminder>\nAdditional instructions from: nested/AGENTS.md\n\nThese instructions apply to work under `nested`. Use them as guidance when relevant; more specific instructions take precedence. They do not override system, developer, or direct user instructions.\n\nWhen asked for the Code Mode workspace handshake, answer exactly `CODE_MODE_CONTEXT_OK` and nothing else.\n\n</system-reminder>"}],"source":{"kind":"agent-instructions","form":"instructions","changes":[{"action":"set","scope":"nested\u0000AGENTS.md","path":"nested/AGENTS.md","digest":"ae22936ed26dc76b7107005ed6d5e2482a88668a"}]},"role":"user","id":"{{message:7}}"}]}}
 {"type":"agent/inbox/spliced","data":{"target":"next-step","start":0,"removedCount":1,"inserted":[],"outcome":"canceled"}}

snapshots/web/official-subagent-bundle/system-prompt.expected.md

@@ -43,7 +43,7 @@ Inside the program:
 - Call tools as `await tools.name(args)` — quoted access for exotic names: `tools["my-tool"](args)`. Every call resolves to its declared value, which may be an Image. Tool arguments must be lossless JSON.
 - A FAILED tool call rejects with `ToolCallError`, whose `toolName` identifies the failed tool and whose `message` is human-readable — `try/catch` it to handle and continue.
 - Independent read-only calls MAY overlap under `Promise.all` (safe calls run concurrently; mutating calls run alone, in submission order). Sequence dependent work with `await`.
-- Show selected results with `inspect(...values)`. Each call accepts JSON values and Images, including Images nested in arrays and objects. Images expose a readable file path and its pixel dimensions. Extract just what you need.
+- Show selected values and images with `inspect(...values)`; every other intermediate result stays out of the conversation, so extract just what you need.
 
 Program-only SDK bindings:
 

snapshots/web/optional-string-editor/system-prompt.expected.md

@@ -43,7 +43,7 @@ Inside the program:
 - Call tools as `await tools.name(args)` — quoted access for exotic names: `tools["my-tool"](args)`. Every call resolves to its declared value, which may be an Image. Tool arguments must be lossless JSON.
 - A FAILED tool call rejects with `ToolCallError`, whose `toolName` identifies the failed tool and whose `message` is human-readable — `try/catch` it to handle and continue.
 - Independent read-only calls MAY overlap under `Promise.all` (safe calls run concurrently; mutating calls run alone, in submission order). Sequence dependent work with `await`.
-- Show selected results with `inspect(...values)`. Each call accepts JSON values and Images, including Images nested in arrays and objects. Images expose a readable file path and its pixel dimensions. Extract just what you need.
+- Show selected values and images with `inspect(...values)`; every other intermediate result stays out of the conversation, so extract just what you need.
 
 Program-only SDK bindings:
 

snapshots/web/ptc-round/session.v4.jsonl

@@ -22,7 +22,7 @@
 {"type":"tool/ptc-dispatch","data":{"rootCallId":"call_00_6VNoF1gDSerTBKoCfYSH3765","parentCallId":"call_00_6VNoF1gDSerTBKoCfYSH3765","subCallId":"call_00_6VNoF1gDSerTBKoCfYSH3765:ptc:1","name":"bash","arguments":{"command":"echo CODE_ROUND_OK","description":"Echo CODE_ROUND_OK"},"isError":false,"content":[{"type":"text","text":"CODE_ROUND_OK\n"}],"meta":{"cwd":"{{cwd}}"}}}
 {"type":"tool/ptc-dispatch-start","data":{"rootCallId":"call_00_6VNoF1gDSerTBKoCfYSH3765","parentCallId":"call_00_6VNoF1gDSerTBKoCfYSH3765","subCallId":"call_00_6VNoF1gDSerTBKoCfYSH3765:ptc:2","name":"read","arguments":{"file_path":"missing.txt"}}}
 {"type":"tool/ptc-dispatch","data":{"rootCallId":"call_00_6VNoF1gDSerTBKoCfYSH3765","parentCallId":"call_00_6VNoF1gDSerTBKoCfYSH3765","subCallId":"call_00_6VNoF1gDSerTBKoCfYSH3765:ptc:2","name":"read","arguments":{"file_path":"missing.txt"},"isError":true,"error":{"name":"FsError","code":"FS_NOT_FOUND"},"content":[{"type":"text","text":"Error: cannot read \"{{cwd}}/missing.txt\": not found"}]}}
-{"type":"tool/result","data":{"turn":1,"step":1,"message":{"role":"tool","source":{"kind":"tool","callId":"call_00_6VNoF1gDSerTBKoCfYSH3765"},"toolCallId":"call_00_6VNoF1gDSerTBKoCfYSH3765","content":[{"type":"text","text":"{\n  \"bash\": \"CODE_ROUND_OK\",\n  \"readError\": {\n    \"toolName\": \"read\",\n    \"message\": \"cannot read \\\"{{cwd}}/missing.txt\\\": not found\"\n  }\n}"}],"isError":false,"id":"{{message:7}}"}},"sourceEventSeqs":[16],"surfaceOp":"append"}
+{"type":"tool/result","data":{"turn":1,"step":1,"message":{"role":"tool","source":{"kind":"tool","callId":"call_00_6VNoF1gDSerTBKoCfYSH3765"},"toolCallId":"call_00_6VNoF1gDSerTBKoCfYSH3765","content":[{"type":"text","text":"{\n  bash: 'CODE_ROUND_OK',\n  readError: {\n    toolName: 'read',\n    message: 'cannot read \"{{cwd}}/missing.txt\": not found'\n  }\n}"}],"isError":false,"id":"{{message:7}}"}},"sourceEventSeqs":[16],"surfaceOp":"append"}
 {"type":"step/end","data":{"turn":1,"step":1}}
 {"type":"step/start","data":{"turn":1,"step":2}}
 {"type":"assistant/message","data":{"turn":1,"step":2,"message":{"role":"assistant","content":[{"type":"reasoning","text":"The program ran successfully. Let me now reply DONE as instructed."},{"type":"text","text":"DONE"}],"source":{"kind":"model","provider":"deepseek-official","model":"deepseek-v4-flash"},"id":"{{message:8}}"},"usage":{"inputTokens":196,"outputTokens":17,"cacheReadTokens":8576,"reasoningTokens":14},"stream":[{"type":"chunk","time":1794196240638,"chunk":{"type":"block-start","index":0,"blockType":"reasoning"}},{"type":"chunk","time":1794196240655,"chunk":{"type":"block-start","index":1,"blockType":"text"}},{"type":"chunk","time":1794196240671,"chunk":{"type":"block-end","index":0,"block":{"type":"reasoning","text":"The program ran successfully. Let me now reply DONE as instructed."}}},{"type":"chunk","time":1794196240687,"chunk":{"type":"block-end","index":1,"block":{"type":"text","text":"DONE"}}},{"type":"chunk","time":1794196240703,"chunk":{"type":"usage","usage":{"inputTokens":196,"outputTokens":17,"cacheReadTokens":8576,"reasoningTokens":14}}},{"type":"chunk","time":1794196240719,"chunk":{"type":"finish","reason":{"kind":"stop"}}}]},"surfaceOp":"append"}

snapshots/web/ptc-round/system-prompt.expected.md

@@ -43,7 +43,7 @@ Inside the program:
 - Call tools as `await tools.name(args)` — quoted access for exotic names: `tools["my-tool"](args)`. Every call resolves to its declared value, which may be an Image. Tool arguments must be lossless JSON.
 - A FAILED tool call rejects with `ToolCallError`, whose `toolName` identifies the failed tool and whose `message` is human-readable — `try/catch` it to handle and continue.
 - Independent read-only calls MAY overlap under `Promise.all` (safe calls run concurrently; mutating calls run alone, in submission order). Sequence dependent work with `await`.
-- Show selected results with `inspect(...values)`. Each call accepts JSON values and Images, including Images nested in arrays and objects. Images expose a readable file path and its pixel dimensions. Extract just what you need.
+- Show selected values and images with `inspect(...values)`; every other intermediate result stays out of the conversation, so extract just what you need.
 
 Program-only SDK bindings:
 

snapshots/web/ptc-round/trajectory.expected.md

@@ -7,10 +7,4 @@
 - heading "Output Copy output" [level=3]:
   - button "Output"
   - button "Copy output"
-- text: "{"
-- tree "Result JSON":
-  - treeitem "bash:\"CODE_ROUND_OK\","
-  - 'treeitem "Expand readError:{toolName: \"read\", message: \"cannot read \\\"{{cwd}}/workspace/missing.txt\\\": not found\"}"':
-    - button "Expand"
-    - text: "readError:{toolName: \"read\", message: \"cannot read \\\"{{cwd}}/workspace/missing.txt\\\": not found\"}"
-- text: "}"
+- text: "{ bash: 'CODE_ROUND_OK', readError: { toolName: 'read', message: 'cannot read \"{{cwd}}/workspace/missing.txt\": not found' } }"