Move exact user-patch watching into app boot and audit activation explicitly.
Wait for CLI fallback HMR and preset subtrees. Join the first fiber-disposal
result in the directory chooser and browser package runner. Keep async
update errors observable without restoring generic Loader rollback.
Restore the current workspace-link and Schemastery manifests and their gate.
Preserve unconditional patch-input cloning, awaited Include initialization,
and durable teardown writes. Update diagnostics, snapshots, bilingual docs,
and the exhaustive vendor divergence ledger.
Validation: focused Loader, watcher, preset, inventory, and cleanup tests;
100% watcher coverage; 32 Web preset cases; real CLI webhook model session;
SDK smokes; 13 headless expected cases; 3 recorded-session replays; build,
typecheck, lint contracts, and all 34 doc-sync gates.
A full Node 24/Python 3.12 Vitest run had 21806 passes and two failures:
a baseline-reproduced experimental CSS import failure and a Python probe
timeout that passed alone. The default Node 26/Python 3.9 run additionally
exposed unsupported Python and FileHandle-GC host failures.
The runtime's user-disabled row ids now come from the committed composition —
composeProfileStack collects them from the user layers it composed — so a
rejected or half-written patch file no longer makes the list disagree with the
running tree or fail to parse. The inventory attributes a row disabled through
a group the user disabled to the user, through the shared userDisablesEntry
walk, and lists a user-layer row the composition left out under the new user
trust instead of builtin.
The installer now decides pnpm's colouring per consumer: the Web host keeps
colours on for the install dialog's terminal, the CLI turns them on only when
stdout is a terminal, and a plain log strips what a colour=always pnpm still
prints. Every chunk carries the argv it came from, so a renderer can label
each pnpm run.
What plugin management does now lives in `dsh-plugin-manager`, a boot-group
package beside `app-boot`: `PluginInstaller` runs pnpm, probes, and judges new
packages over the profile on disk, and `PluginManager` adds every operation
over the booted tree, each refusal a `PluginOperationError` with a `plugins/*`
code and no Remote protocol attached. The manager reads the profile runtime,
the preset roster, and the running-agent count through per-call readers, and
the roster only as `PresetLayers`, so the package depends on neither the
roster nor the agent registry.
`dsh-host-plugin-manager` keeps `pluginManager` and the `plugins` Remote as a
relay: one Remote method per manager method, and one exhaustive switch
turning a failure into the Remote error of the same code (`plugins/bad-request`
into `gateway/bad-request`). Its `./types` re-exports the manager's types, so
`api-remotes` and the client import one vocabulary unchanged.
`dsh plugin add` and `remove` now run the installer instead of a bare pnpm
forwarder: every new package is probed, a package that declares neither a
bundle nor a plugin module is removed again with its reason, and newly
installed bundles are enabled as before. Other pnpm verbs still forward.
`ProfileRuntime.recompose({ reloadBundles: true })` replaced its profile
before the Loader had accepted the update, so a rejected recomposition left
`current`, `layers`, and `originOf` describing a stack that never mounted.
The user patch watchers took a second path through the launcher's
`composeLive`, which wrote the new conflicts into the failure registry
before the update and kept composing from the runtime's stale profile.
Conflict records shared the registry with row failures under a fabricated
entry id and a group id that meant three different things, so the plugin
inventory attributed a conflict to the layer that won the id, and a bundle
disabled after failing kept its import and apply records forever.
The runtime now holds one committed composition — profile, stack ownership,
conflicts — and publishes a candidate only after `entry.update` resolves;
`originOf` reads the committed owners instead of recomputing them. The
watchers call `recompose` through a `reapply` callback, so every live change
goes through the one entry point. Conflicts live on the runtime with their
messages and never enter `pluginFailures`, whose records name rows that
reached the Loader; the inventory lists them under the layer that lost. A
contained group clears its rows' records when it unmounts.
Adopt owned row ids in the plugin manager and inventory: a row view carries
`rowId` (the declared id a patch or row action targets) instead of the
withdrawn `originalId`, probe rows are listed under their declared ids, a
conflict record is attributed to the bundle it names, and the manager's
test boot records conflicts the way the launcher does.
An external bundle keeps the ids its patch declares. composeProfileStack
decides ownership before anything mounts: built-in and boot-staged layers
claim first and a duplicate among them fails the boot, a contained bundle
that collides is left out whole, a user-layer insert of a taken id is
dropped, and every row left out is a `conflict` record in pluginFailures,
printed at boot and replaced on each recomposition. Boot, live reloads, and
--dump-config compose through the same function, so the dump shows the
contained groups the tree mounts.
The prefix made a user patch, a third-party manager's disabled rows, and a
bundle's own `!!js` id comparison miss silently; the note records both cases.
Add `dsh-host-plugin-manager`: the `pluginManager` service and the
`plugins` Remote that install, uninstall, enable, disable, and retry
bundles over the booted profile, add and remove rows in the profile's
or an agent preset's user layer, report what a package's services keep
alive, and fold the manifest, the probe record, and the live tree into
one view per package. pnpm runs the way the CLI runs it, streaming its
output as `plugins/install-log`; every change emits `plugins/changed`,
and api-remotes forwards both. Enabling is the Loader's own
transaction: a rejected recomposition restores the layer list, and a
runtime recomposition records contained rows through
`recordContainedStates` because the boot audit does not run again;
`ContainedGroup.create` now records a row that resolved pending rather
than clearing it.
Give every agent preset a user patch layer: `cordis.patch.yml` beside
a locally authored composition, or alone in `.agent-presets/<id>/`
for a shipped one. Discovery attaches and judges it, the mount applies
it as the include's runtime patches, standing generations follow the
layer's content and return to a retired generation when it comes back,
the composition inventory reports each row's `source` and
`disabledBy`, a copy carries the layer, and `removeOverlay` deletes it.
`dsh-global-tool-mask` is `tools.restrict()` as a scope-only row for
such a layer.
Add `dsh-patch-file` as the one home of the patch-list format: the
parser moved out of app-boot, plus a comment-preserving key-level
writer with lock, atomic replace, and readback. app-boot exports
`layerTrust`, `bundleLayerPatches`, and `recordContainedStates`, and
`ProfileRuntime` carries the install anchor.
Mount every external bundle layer as a cordis:contained-group entry
`bundle/<package>` whose rows carry a `<package>/<id>` prefix, so one
community bundle that no longer loads records a failure in the root
`pluginFailures` registry instead of rejecting the whole Loader
transaction, and two bundles may declare the same row id. Classify
layers by origin (`builtin` for the template bundle and
`dsh.profile.firstParty`, `external` for profile dependencies) and let
`dsh.bundle.stage: boot` or `dsh.profile.stages` opt a bundle out of
isolation when it provides a service built-in rows inject; a built-in
row left pending beside an isolated bundle still fails boot and names
the bundle.
Keep `installFailLoud` for the boot transaction only and install
runtime guards afterwards: an unhandled rejection is reported and
contained, an uncaught exception is reported and exits. Report nested
fiber failures under built-in entries as advisory lines.
Provide `ctx.profileRuntime` from the launcher (layers with trust and
stage, row origin, user-disabled rows, recomposition through the root
include) and probe installed packages in a child process
(`probePackage`, cached under `profiles/<name>/.dsh-plugins/`). Split
installation from enablement in `reconcileInstalledBundles`; the CLI
keeps install-means-enable. The plugin inventory now reports `trust`,
`package`, `disabledBy`, and `failure` per row and lists rows only the
failure registry knows.
The package exported six functions, four of them shaped by one SDK's
transport each: a dispatcher factory, a `node:http` agent factory, a
proxy-URL lookup, and a policy accessor. Review asked whether the call
sites could converge instead of the package growing an export per SDK.
They could, and each removal took a whole shape with it:
- The OTLP exporter moves to the SDK's `fetch` delegate, retiring
`createNodeHttpAgent`. Its Node-version floor goes too: `proxyEnv` on
an `http.Agent` needs 22.21 or 24.5, inside the engines range, so
telemetry was direct on 22.19, 22.20, and 24.0-24.4. The cost is
`compression`, a Node-transport option; the plugin now refuses it,
`keepAlive`, and `httpAgentOptions` at load instead of ignoring them.
- `web-fetch-http` builds its own address-pinning agent under an
annotated `proxy-exempt:` exemption, retiring `createDispatcher`.
Pinning is per-request state a process-wide dispatcher cannot hold.
- E2B reads `route.proxy`, retiring `proxyUrlFor`.
What remains is `installProxyFromEnvironment`, `proxyRouteFor`,
`proxyEnvironmentForChild`, and `clearedProxyEnv` — one per way a caller
can need the policy. Installation absorbs resolution and diagnostic
reporting, which no caller needed apart.
`proxyRouteFor` also closes a defect the old accessor made expressible:
`web-fetch-http` read the policy to decide whether to pin, then read it
again to build a transport, so an unmount between the two returned a
direct, unpinned agent for a URL the first read had cleared as proxied.
A route carries the answer and the transport that answer assumed.
Every egress spec now installs through `installProxyFromEnvironment`, so
no test asserts a policy object a real launch could not produce.
Node's built-in fetch ignores HTTP_PROXY, so every harness request connected
directly regardless of what the user exported. Resolve one policy from the
launch environment and install it as undici's global dispatcher, then wire the
four surfaces a global dispatcher cannot reach: web_fetch's pinned transport,
the OTLP exporter's node:http agent, the E2B SDK's own proxy option, and the
environment a child process or worker thread is given.
Each outbound call site carries an egress test that drives its real code path
through a fake proxy; that measurement is what found the OTLP and E2B gaps.
Record sdk-minimal as the narrow repository-owned exception to base-first profile composition: callers still launch only dsh and cannot provide an arbitrary Cordis tree, while the shipped bundle may own a complete explicit roster. Cross-link the launcher, profile-bundle, Python-runtime, minimal-agent, snapshot, and telemetry decisions; the supersession audit keeps each older note active because its remaining rationale is independent.
Update the CLI, architecture, Python tutorial/reference, example, runtime-wheel reference, and bundle documentation. The docs distinguish the full sdk profile from sdk-minimal, explain explicit-home/plugin/patch customization, state the minimal permission and persistence choices, and retain the separately packaged web profile and frontend assets for direct dsh use.
Correct dsh-base descriptions to cover base-backed profiles, make SDK startup configuration visible in the generated config catalog, add sdk-minimal to the module graph, and regenerate the base-composition graph. English and Chinese pairs are re-recorded at the exact reviewed contents.
Resolve packaged profile proxies with Node ESM import conditions from each package installation, and fail loud when an explicit runtime export or legacy main entry is missing. Serialize the shared profile fallback under the existing cross-process writer lock so concurrent dsh processes cannot observe partial proxies; either carrier now replaces the other carrier’s managed entry without manual cleanup.
Give Python initialize its own 10-second default bound and name the selected profile in timeout diagnostics, while leaving ordinary agent turns unbounded by default. Package the dynamically resolved web frontend and skill-badge assets so the runtime wheel’s normal dsh profiles do not depend on pkg static-discovery accidents.
Rewrite the root launch rule and every active stale SDK-runtime note to the shipped dsh profile architecture in both languages. Focused tests prove import-only and transitive package exports, lock contention, cross-carrier transitions, missing-entry failures, asset inventory, and bounded initialization.
Move the shared module-reload policy into dsh-base by inserting its HMR row disabled, then remove the redundant disabled overrides from Web, headless, SDK, and ACP. No shipped profile enables server module reload; live profile patch watching continues through the launcher-owned config-only fallback, and browser client HMR remains a separate mechanism.
A later profile layer can opt into source-module reload explicitly with disabled: false while retaining the base root configuration. Composition tests cover every shipped mode and the explicit enable path, and the bundle references plus launcher Agent Notes document the resulting ownership and safety rationale.
Replace the TypeScript SDK's public arbitrary command/argv launch surface with the same-version dsh CLI, a named profile, ordered per-launch patches, optional process cwd, and explicit Harness home selection. Installed consumers use the built CLI; clean source checkouts use the package's src/bin.ts through an absolute tsx/esm loader and a source-only patch that omits build-generated Typert loading.
Materialize explicit or inherited environments at spawn time, keep profile-internal patches below caller patches, and resolve every caller-relative path before the child starts. The SDK subagent provider validates its optional CLI and patch files at plugin load and requires an isolated absolute child home.
Treat JSON-RPC initialize as the Loader-owned readiness point: Loader settlement joins entry imports, fiber lifecycle work, and synchronous effect registration, so the server needs no scheduler tick. A delayed profile entry registers a private adapter before initialize resolves, proving caller-supplied plugin routes are visible without fallback.
Update sdk-app ownership, server diagnostics, TypeScript SDK examples, nested-loader coverage, and session-upload replay layering together. The following commit contains only the regenerated SDK transcripts, keeping this API and lifecycle change directly reviewable.
Yichen Jiang