This pull request lands the tool; switching the shipped defaults is a
separate pull request stacked on it. Every shipped `tool-fs` row states
`applyPatch: false`, the `fs-observation-policy` row stays mounted, the
write, edit, and apply_patch guidance names the policy, and every recorded
expectation outside the apply-patch scenarios returns to the master
baseline. The upgrade guide, the parity exception, and the optional
classification of the policy package move to the stacked change.
Evaluation on the shipped Web composition decided the defaults: the
`fs-observation-policy` row in dsh-base ships `disabled: true`, so write,
edit, and apply_patch mutate unconditionally in every profile; the Web
`standard` preset lists apply_patch; and the tool-fs `applyPatchGuard` switch
is gone, apply_patch always consulting the fs/* intent waterfalls like the
other mutating tools.
The write and edit guidance stops citing the policy as the default, and the
apply_patch guidance asks for enough hunk context instead of a prior read.
The two recordings that capture a policy refusal (fs-policy-reject and
apply-patch-rejected) re-enable the row through a per-scenario
runtime.cordis.yml; every other sidecar follows the prompt change, and the
Web standard-preset recordings gain the apply_patch schema. An upgrade guide
records the default change.
`tool-fs.applyPatchGuard` selects the mutation guards of apply_patch:
`observation` consults the mounted observation policy per operation like
write and edit; `none`, the default, applies hunks unconditionally on
context match alone as the Codex tool does, keeps publishing fs/observed,
and replaces the read-first guidance with a context-matching reminder.
In 2,226 recorded DS-Bench and DeepSWE rollouts with the policy on, read-first
and stale-version refusals were 29–41% of apply_patch failures although hunk
context matching already rejects a divergent file. The recorded apply-patch
scenarios pin `observation`, under which they were captured.
Across 2,226 recorded DS-Bench and DeepSWE rollouts, 14% of apply_patch
calls arrived as {"input": ...} and were rejected for the missing `patch`
property: models trained on the Codex tool send its parameter name
unprompted. The schema now uses `input`; the executor request keeps its
internal `patch` field. Recorded apply-patch sessions, the schema sidecar,
and the catalogs follow.
Takes master's re-measured npm install-layout budget, which covers this
branch's graph, and regenerates the Cordis API, event matrix, client slot,
and tool catalogs on the merged tree.
Resolves the headless snapshot runner (keeps both the readonly-patch-directory
and cua-script-guidance workspace setups), regenerates the client slot and tool
catalogs with their pairing record, adds the apply_patch tool title to the five
locales master introduced, and refreshes the apply-patch system-prompt pin for
master's subagent guidance change.
`tool-fs.applyPatch` now defaults to false instead of deriving from
`write && edit`, so every shipped composition keeps the model request it
had before the tool existed. The `tool-fs` rows that enable `edit`
(dsh-base and the Web standard, ptc, and cordis presets) state
`applyPatch: false` explicitly; changing that one line adds the tool.
The apply-patch snapshot composition enables the switch, the tool catalog
mounts tool-fs with it on so the schema stays documented, and every other
recorded expectation returns to the master baseline.
Master replaced the creation and plugin recommendation bundles with one
personalized recommendations bundle.
- The CLI manifest keeps master's bundle and the forge bundle.
- The plugin page expectation ranks the forge bundle before the held-back
personalized recommendations bundle.
- The experimental README pairing is recorded again.
The blocking ask_user_question description did not explain that an answer
item with empty selected and no custom is a user skip, so models re-asked
the same question after the user clicked Skip.
Refs #7134
* feat(experimental): ship the two card experiments as one Personalized recommendations bundle
@deepseek-ai/dsh-personalized-recommendations inserts the action card
service, its browser half, Personalized pages and Plugin recommendations
with their card bodies under the existing row ids. As the only layer that
declares those ids, the Plugins page addresses and switches each row,
including the two shared card rows that were read-only while two bundles
inserted them.
The two experiment packages stop being bundles; the new bundle takes their
place in the official lists and the dsh installation, with the personalized
pages icon, localized title and description, and a feature preview of both
cards. The two feature rows describe what they do on the bundle page.
* fix(experimental): drop the model reason line from plugin recommendation cards
The reason follows the language of the user's question while the catalog
description below it follows the interface language, so the two lines could
differ in language and largely repeat each other; the answer already gives
the reason. The card keeps the catalog title and description, both in the
interface language, as its details target.
* test(web): refresh the card fixtures for the Personalized recommendations bundle
Both card replays enable the new bundle, so each Session lists both card
skills, and each skill's catalog lists Personalized recommendations in place
of the two former bundles. The plugin card goldens lose the reason line.
---------
Co-authored-by: ZiyaZhang <[email protected]>
The parent branch now carries master through #6426 again.
- Master moved the optional, experimental, and plugin page bundle lists
into @deepseek-ai/dsh-official-bundles; the forge bundle joins the
optional and experimental lists there.
- The CLI manifest, release families, and plugin page expectations keep
both sides; the forge bundle ranks before the held-back recommendation
bundles.
- The lockfile, the event matrix, and the experimental persistence
catalog are regenerated, and the capability-seams and experimental
README pairings are recorded again.
Takes the card packages' new names and master up to e115cc98d65. The plugin
recommendation packages and their bundle rows now name
@deepseek-ai/dsh-action-cards and @deepseek-ai/dsh-client-ui-action-cards;
generated catalogs are regenerated, and both card Web fixtures record the
renamed bundles in their skills' plugin catalogs.
The four packages keep their directories under packages/experimental and
are registered as experimental name exceptions:
- @deepseek-ai/dsh-action-cards
- @deepseek-ai/dsh-client-ui-action-cards
- @deepseek-ai/dsh-creation-recommendations
- @deepseek-ai/dsh-client-ui-creation-recommendations
Loader row ids, plugin record names, message source kinds, the Remote
service name and locale namespaces are unchanged. Generated catalogs are
regenerated, and the page card Web fixture records the renamed bundle in
the skill's plugin catalog.
Plugin recommendations now proposes through the shared propose_action tool
instead of a dsh-action answer block. The Host registers the
enable_official_plugin kind with the action card service: the
official-plugins skill, details holding the plugin name, the Enable verb
with global access, and the enabled, already-enabled and failed phases,
whose data carries the application outcome and the management error code.
The package drops its own projection, Remote act, status notice and
persisted record; it keeps the describe Remote for card text. The bundle
inserts the shared card rows beside its own, as Personalized pages does.
The browser half registers a card body in the shared frame that keeps the
catalog text cache, the reserved loading lines, the single details target
and the Enable and Not now footer. The skill's submission section now asks
for one propose_action call.
action-cards owns the propose_action tool, the actionCards projection of
proposals and decisions, the act Remote and the card status notice; card
kinds register an action name, skill, details schema, verbs and handler.
The tool is admitted per Agent only after a registered kind's skill loads.
client-ui-action-cards draws the card frame below the completed answer,
lets kinds register a body per action and adds the propose_action
process row.
The parent branch now carries master through #6426, including the
agent-controller revert.
- Package manifests keep both sides: the ./types and ./locale/*.json
exports of the forge packages, and the Web app's forge bundle and
message quotes dependencies.
- The type manifest drops the removed agent-controller entries and keeps
the forge panel entries.
- The event matrix and the capability-seams pairing are regenerated.
- The forge bundle stays on the main plugin list and now ranks before the
held-back developer tools; the plugin page expectations follow.
The proposal is a fenced dsh-action JSON object at the end of the final
answer instead of a suggest_action tool call. The projection reads the
last valid block of an assistant message, keeps one card per turn, and
marks a proposal written while queuedMessageLimit user messages are
queued as suppressed; the chat hides dsh- fences from rendered answers.
The Web replay fixture, tool catalog and READMEs follow the toolless
mechanism.
Follow the tool executor refactor: the tool/ptc-dispatch event type now comes
from dsh-tool-executor-ptc, and the tests invoke tools through dsh-tool-testkit.
Regenerate the doc graphs, tool catalog and persistence schema after the merge
and re-record their translation pairing.
Drop the preflight sandbox fence. It derived writable roots in the local
process, so under the SSH provider it threw on remote additionalWritableRoots
and compared remote targets against local temp roots; write and edit never
fenced in the tool layer. The provider now refuses a denied operation at
commit, and a commit whose first operation failed without any change rejects
like a preflight refusal so the sandbox escalation path keeps its error code.
Classify FS_ABORTED as unknown: a cancelled remote request rejects while the
provider may still be writing, so no stat comparison can settle it. Document
that a cancelled call receives the registry's abort result like every tool.
Store no diff for creations, matching write, and bound the move source basis
by diffBasisMaxBytes so persisted metadata never holds whole files;
FsMoveOutcome.sourceBefore becomes nullable. Omit deletions from the call
preview and from result diffs without a basis instead of rendering them as
empty creations, and count distinct files in the preview title.
Reject a NUL byte beyond the read sampling window in preflight, name targets
by display path in atomic-write diagnostics, align the apply_patch guidance
with the edit wording and mention delete and move, read the stopped flag and
applied diffs tool-agnostically in the client models, skip the permission-bit
snapshot under root, list fs/delete-intent in the tool catalog, and refresh
the fs-patch description, group README, fs-sandbox summary, and ui-tool README.
Generated worktree names now join the dsh- prefix, the first eight
hexadecimal digits of the UUID in the Session id, and four random
hexadecimal digits, such as dsh-6b1f02c4-9a3e. A Session id without a
UUID gets eight random hexadecimal digits after the prefix. The worktree
runtime exports generatedWorktreeName, uses it for omitted
create_worktree names, and defaults namePrefix to dsh-. The forge panel
names its Session worktrees with the same function and defaults
worktreePrefix to dsh-, so checkouts no longer nest under a dsh/
directory.
The create_worktree name parameter asks for a few lowercase words joined
by hyphens that describe the task, and the tool description drops the
defaults that its parameters already state.
Regenerate config/tool/module/event catalogs, Cordis API and client slot catalogs; re-record reviewed bilingual hashes instead of merging generated bytes.
Accept deleted filesystem invariant companions. Master has no replacement event-invariant registration; keep delete-intent in typed events, policy guards, and behavior tests.
Adopt master workspace-changes intent waterfalls and tool-execution recorder association, add a prepended delete-intent listener that delegates before capture, remove argument parsing and the fs-patch dependency, and retain the real-tool Review matrix.
Keep master filesystem mutation-result metadata and tool switches alongside opt-in apply_patch. Await current-directory resolution in plan and no-follow preflight; store patch diffs as processPath. Adapt native/PTC fixtures to the supported modes.
Preserve client patch diffs and stopped states with master partial arguments and absolute navigation. Use master's rewritten ProviderEditor, including its remaining repeated model validation.
Regenerate the lockfile with pnpm install and align the new fs-patch package version. Clean retired package build outputs. Update exact rescope exceptions for verified preset and event identifiers from master; old package names remain rejected.
sjw1231