* fix(#2875): stage user artifacts durably across install wipes (#1874-F19) preserveUserArtifacts held user files only in an in-memory Map across the wipe, so any process death between preserve and restore lost them outright. Seven call sites, not the four the issue records. Three of them never called the helper at all - they open-coded the same read/wipe/write - so searching for callers under-counted by construction; the extra sites were found by sweeping for the pattern instead. The worst is the mainline install path, where the crash window spans the entire gsd-core tree copy rather than a single rmSync. Adds src/user-artifact-staging.cts: durable on-disk staging with a record written after the copies land as the commit point, plus recovery of orphaned batches on the next run - without recovery the staged bytes survive but the user's file is still gone, which would pass its own test while delivering nothing. Routes copyPreservingSymlink through installFs() so staging cannot bypass the install fs seam, and reunites its symlink-safety docblock with the function it documents. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> * docs(#2875): amend ADR-3574 with four claims disproved by implementation Implementing Phase 6 disproved four statements the ADR rests on. The central decision - no single materializer - is unaffected and stands. Corrected: decision 3 was already satisfied, so nothing was extracted; the agents-bypass runtime set omitted claude, kilo and opencode, and closing it needed three new pieces of descriptor contract rather than proceeding on its own terms; three of the four blockers the layout comment names were already stale; and F19 is seven call sites, not four. Records the generalizable lesson: the defect is the pattern of holding user data in memory across a wipe, not the helper, so searching for callers of the helper under-counts by construction. Also resolves the ADR's open question on USER_OWNED_ARTIFACTS membership, and notes that copyPreservingSymlink needed routing through the install fs seam before it could be reused. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> * fix(#2875): close dangling-symlink blind spot and harden staging recovery An adversarial review found the F19 staging work shipped red and unsafe. Root cause, shared by two arbitrary-write findings: hasExistingSymlinkBetween missed dangling symlinks in both its root check and its per-segment walk, because it probed with existsSync, which is false for a link whose target does not exist. Fixing only the new module would have reused a guard that was itself blind. This guard protects the whole install tree. Recovery no longer throws: it degrades per entry and per file, so one bad batch cannot block the others. Previously an unrecoverable entry propagated out of the first statement of install and uninstall, before the cleanup that would have removed it - wedging the installer permanently. Partial fs adapters now throw on any omitted method instead of silently reaching the real filesystem, closing the trap that let a test poison list pass while real IO happened. Staged names must be flat, recovery refuses a dangling destination symlink, and a batch whose recovery genuinely failed is no longer swept - it was discarding the only durable copy of the file it had just failed to restore. Replaces three tests that could not fail, including the one labelled negative proof. Known limitation, documented not closed: concurrent installs sharing a staging key can still lose a batch. A real fix needs a cross-process lock. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> * enh(#2875): make the descriptor authoritative for the agents kind Deletes the inline agent-staging loop in bin/install.js and the _DESCRIPTOR_AGENTS_RUNTIMES set, so every runtime materializes agents from its capability descriptor instead of an inline hostBehaviors dispatch. Closing it needed three pieces of contract the descriptor pipeline never had, all reducible to one missing input - per-agent resolution context: a frontmatter-extensions step for claude's effort and disallowedTools, per-agent model-override resolution for kilo and opencode, and a named branding converter for hermes, whose rewrite data was already declared. Seven runtimes were on the loop, not the six the design recorded - kimi-code was found by a golden fixture, not by analysis. claude-local and kimi-code both silently lost their agents mid-change; the fixtures caught both and the cause was fixed rather than the fixtures regenerated. A parity harness gates the migration: both pipelines over identical inputs, byte-identical output including filenames, per runtime. It is demonstrated red before being trusted. Surface and install paths converge for all seven, which also fixes surface previously writing no agents for these runtimes. Codex's config.toml strip stays put - it mutates host config, which no descriptor kind models. Also routes install-model-override-resolver and install-effort-resolver through the install fs seam. Both leaked real filesystem IO from the install call tree; the stricter adapter is what exposed them. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> * docs(#2875): record the agents-descriptor migration and correct the ADR count The _DESCRIPTOR_AGENTS_RUNTIMES allow-list no longer exists, so the host integration guide told readers to join a set that is gone. Replaces that with what is now true - declare an agents entry and it installs, on the surface path as well as install - and points anyone needing a per-agent transform at the three extension points rather than at a new inline branch. Corrects the ADR amendment: seven runtimes were on the inline loop, not six. kimi-code was found by a golden fixture going red, not by reading. That is the third short count this phase, all from enumerating by symbol or set membership when the thing that matters is a behavior. Adds the Changed changeset for the surface-path convergence. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> * docs(#2875): amend ADR-2866 - claude global always wrote agents on disk The claude row's global=[skills] described what capability.json declared, not what the installer wrote. bin/install.js's inline agent-staging loop was never scope-gated and never consulted the descriptor, so a claude --global install has always written agents/gsd-*.md. Phase 6 closes the gap by deleting that loop and declaring agents on claude's descriptor at global scope. On-disk bytes are unchanged - the golden fixtures did not move, which is the evidence that the descriptor, not the installer, was incomplete. #2218 is unaffected: agents are not trigger-bearing, so the wider row does not introduce a new shadowing case. Records the warning that an incomplete descriptor is invisible while a second code path silently does its work, and only surfaces when the two are forced into agreement. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> * fix(#2875): close review findings across staging, agents and the parity harness Two independent reviews of this branch found defects the local gates missed. Security: a dangling symlink at a migration destination allowed writing outside configDir - the same class this change claimed to close, missed at the terminal write of the flow being added. The staging-root resolver threw as the first statement of install and uninstall, so a hostile symlink bricked both, and symlinked-configDir users lost uninstall as well as install; it now degrades instead of aborting. Recovery gained a source-side symlink check and now refuses a relative destDir, which resolved against cwd. Converter dispatch gained a runtime allowlist - lint-time validation stopped mattering once this branch promoted that dispatch from the surface path to real installs. Correctness: claude --local --minimal exited 1 because the minimal profile legitimately yields zero agents and the new path treated that as a failure. cline --local silently lost its agents - its descriptor declared none while the deleted loop wrote them unconditionally. The agents prune was widened to any gsd-* entry and destroyed user files it never owned. The parity harness, on which the migration's safety argument rested, drove a synthetic registry and never byte-compared the shipped descriptors; two of its trap rows could not fail. It now drives the real registry across 13 runtime-scope rows including kimi-code and cline-local, and its red-proof is demonstrated by corrupting a live capability.json. Three goldens that had encoded the cline regression as expected behavior were corrected. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> * fix(#2875): close findings from both mandated review engines /security-review found the staging source-side walk honouring GSD_ALLOW_SYMLINKED_DEST, an opt-in documented as relaxing only the write destination. A symlinked files/ component dereferenced because copyPreservingSymlink lstats the leaf only, so an intermediate link is followed. The source walk no longer honours the opt-in; the destination check still does. /code-review spec axis found this branch had reintroduced its own bug: migrateLegacyDevPreferencesToSkill's new symlink refusal threw unguarded after the legacy dir was wiped and before the staged batch was restored, so a planted symlink bricked uninstall permanently and orphaned the batch. Refusal kept, abort removed. kimi-code local silently lost its agents, the same class as the cline bug, and the parity harness recorded that exclusion as intentional - the third test in this branch to pin a regression as correct. --minimal now creates an empty agents/ dir that never existed. Behaviour restored rather than softening the changeset, so its byte-identical claim stays true. Standards axis: try/finally removed from twelve test bodies, fast-check properties added for parseOwnerPid, boundary coverage at the grace window and the ancestor-probe depth, a parity assertion for the staging-root helper duplicated across two files, and the 8-deep config walk deduplicated. Records 60-review.json with every finding and disposition from five passes, including the smells left unfixed and why. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> * fix(#2875): prune stale agents unconditionally in minimal mode The previous round stopped an empty agents/ directory being created when the resolved profile yields no agents. That was implemented by skipping the agents kind entirely, which also skipped its stale-agent prune - so a full to minimal downgrade left stale gsd-* agents behind. The deleted inline loop pruned unconditionally and only skipped writing. Those are three separate conditions, not one: prune always, write only when there is something to write, create the directory only when writing. Both call sites now run _removeGsdEntries before the empty-staged early exit. The symlink-escape guard moved with it, since the prune also touches dest. Codex .toml agents and the config.toml stanzas are cleaned again, and user-owned agents are still preserved. The agents/ directory is left in place after a prune empties it, matching every sibling kind - none of them remove the destination directory itself. Golden fixtures confirmed byte-identical: the prune is a no-op on a fresh install, so fixture generation is unaffected. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> * docs(#2875): document interrupted-install recovery for user-owned files The durable-staging fix is invisible to the user it protects. Someone whose install died mid-flight has no way to know USER-PROFILE.md was staged before the delete, that the next run restores it, or that recovery happens at the start of that run rather than in the background. Written as the task the user has - finish the interrupted command - rather than as a description of the mechanism, and states what it will not do: overwrite a file already present, or touch staging belonging to another install still running. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> * chore(#2875): backfill changeset pr number Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> * test(#2875): assert the J8 model override without building a regex CodeQL flagged incomplete string escaping: the assertion interpolated the override value into a RegExp while escaping only forward slashes, which is meaningless in a constructor, leaving real metacharacters unescaped. The failure direction was the dangerous one - a metacharacter would have made the match more permissive, so the row would pass when it should fail. That matters here because J8 exists precisely because an earlier revision was a tautology; the rewrite reintroduced a different way for the same assertion to stop discriminating. Replaced with a line-wise exact match, so no regex is constructed at all. Swept the other test files this branch adds; no sibling instances. lint:ci passed on the original - lint-no-adhoc-regex-escape matches a full metachar-escape copy, so a single slash replace slipped under it. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> --------- Co-authored-by: sim <sim@local> Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
12 KiB
How to add or update a host's integration capabilities
This guide is for GSD maintainers adding a new host CLI, or updating an existing host's
host-integration axes (ADR-1239 Phase A). It covers the documentation-sourcing rule, the
nine runtime.hostIntegration axes, the undocumented sentinel, and how to validate.
The governing rule for this whole process: every axis value must come from the host's own
authoritative documentation. Never infer, guess, or assume. Where the docs do not state an axis,
record the explicit undocumented sentinel — not a plausible default. The reference matrix
(docs/reference/host-integration-capability-matrix.md) is the source of truth, and every value in
it carries a citation and an evidence quote.
1. Find the host's authoritative documentation
In order of preference:
- Context7 —
resolve-library-idfor the host, thenquery-docsfor "plugins / subagents / hooks / commands / MCP / model API". - Official dev docs / source repo — the host's documentation site or GitHub repo (plugin API, agents, hooks, MCP, command authoring).
Capture the exact source (Context7 library id + query, or the doc URL) and a short verbatim quote for each value you determine. You will paste these into the matrix in step 4.
2. Determine each of the nine axes from the docs
Read the docs and map them to the closed vocabulary. Do not pick a value unless a source states it.
| Axis | What to look for in the docs |
|---|---|
embeddingMode |
An in-process programmatic plugin/extension API (imperative) vs. configuration files only (declarative). |
commandSurface |
How custom commands are authored/invoked: slash-file (.md), slash-toml, slash-programmatic, palette, prose-only. |
dispatch |
Sub-agent delegation: namedDispatch, nested, maxDepth (int; -1 = documented-unbounded), background, subagentToolkit (full/read-only). |
modelMode |
A programmatic model request/provider API (active) vs. instruction/per-agent-field only (passive). |
hookBus |
The host fires lifecycle events a plugin subscribes to (host), an extension host owns the bus (engine), or no bus (none). Independent of hooksSurface — e.g. opencode has hooksSurface: none but hookBus: host. |
stateIO |
filesystem, sandboxed-storage (web IDE, no arbitrary FS), or session-log-append. |
transport |
mcp (native MCP support) vs. native-extension (MCP needs a community extension). |
runtime |
The plugin/extension runtime: node, bun, sandboxed-web, python, go, rust, electron, other. |
effortSurface |
How reasoning effort reaches the host: argv (a flag on the host's own invocation) or none (no reasoning-effort mechanism). Added by #2481. There is deliberately no config-file member — do not invent one; use undocumented when the host's docs state no reasoning setting. |
3. Write the runtime.hostIntegration block
In capabilities/<id>/capability.json, inside the runtime object, add (or edit) the block. Use a
documented closed-vocabulary value, or the literal string "undocumented" for any axis the docs do
not state:
"hostIntegration": {
"embeddingMode": "declarative",
"commandSurface": "slash-file",
"dispatch": { "namedDispatch": true, "nested": false, "maxDepth": 1, "background": false, "subagentToolkit": "undocumented" },
"modelMode": "passive",
"hookBus": "host",
"stateIO": "filesystem",
"transport": "mcp",
"runtime": "node"
}
When to use undocumented: only when you searched and the host's docs genuinely do not state the
axis. It validates, but negotiateHostCapabilities fail-closes on it (degrades to the most
restrictive known value) — so it is always safe and never a silent capability claim. A dispatch
boolean or maxDepth may also be "undocumented".
Do not conflate the orthogonal axes: commandStyle (GSD's emission style) is not
commandSurface (the host's surface type); the hookEvents dialect is not hookBus (bus
ownership); runtimeCompat (which features run on a host) is independent of these runtime→engine
axes.
4. Record the citations in the reference matrix
Add (or update) the host's section in docs/reference/host-integration-capability-matrix.md with a
row per axis: Axis | Value | Source | Evidence. For an undocumented value, put the search trail
in the Source column. This file is the deployment source of truth — a value without a citation here
is not allowed.
5. Validate
npm run build:lib
npm run gen:capability-registry # validateRuntimeBody runs on every descriptor
gen:capability-registry must succeed with zero errors. The validator
(gsd-core/bin/lib/capability-validator.cjs) rejects out-of-vocabulary values, malformed dispatch
structs, and reserved keys (__proto__/constructor/prototype).
Then run the host-integration tests and the full cross-platform suite:
node --test tests/host-integration-descriptors.test.cjs # asserts every descriptor validates + profiles
gsd-test-both # Mac + Linux Docker (run before any PR)
6. If you need a vocabulary value that does not exist yet
The vocabulary is intentionally closed (ADR-857 Decision 8): a genuinely new host shape requires
a first-party primitive, reviewed. To add one (e.g. a new runtime kind):
- Add the value to the relevant axis in
HOST_INTEGRATION_AXESinsrc/host-integration.cts. - Add the same value to the matching
VALID_*set incapability-validator.cjs.
The parity guard (tests/host-integration-validator-parity.test.cjs) fails if these two drift, so
they must be updated together. Document the new value's meaning in the matrix legend.
7. Fold an already-hardcoded host into the interface (worked example: claude)
Sections 1–6 cover a green-field host (pi, antigravity — a fresh descriptor + reference
binding). This section covers the other case: a host that already has a real production install
driven by scattered runtime === '<id>' string-equality branches in bin/install.js, which you want
to move onto the Host-Integration Interface without changing a single installed byte. claude
(the tier-1 reference host, #2086) is the worked example.
The pattern is byte-parity-safe by construction — each string check becomes a descriptor lookup that yields the same truth value, so behavior is unchanged and only the brittle coupling is removed:
-
Inventory the branches. Find every
runtime === '<id>'/runtime !== '<id>'inbin/install.jsfor the host (grep -nE "runtime\s*[!=]==\s*'claude'"). Each is a host behavior encoded as a string comparison rather than a declared capability. -
Declare the behaviors on the descriptor. Add a
runtime.hostBehaviorsobject to the host'scapability.json. Each key names one behavior the branches gated on — e.g. forclaude:permissionsSchema: "claude",settingsFileByScope: { local: "settings.local.json", global: "settings.json" },sourceMarkerFile: ".gsd-source",agentFrontmatterExtensions: ["effort"],localInstallStyle: "legacy-flat",authorsCanonicalWorkflow: true,ownsClaudePaths: true,nativeModelAliases: true,skillsGlobalOnboarding: true,attributionSource: "settings-json-commit". The validator (validateRuntimeBody) is lenient toward these host-behavior keys; they carry install policy, not the closed negotiated axes. -
Replace each branch with a descriptor read.
bin/install.jsexposes a_hostBehaviors(runtime)helper (reads_capabilityRegistry.runtimes[runtime].runtime.hostBehaviors,{}if absent). Rewriteif (runtime === 'claude')→if (_hostBehaviors(runtime).permissionsSchema === 'claude'), andif (runtime !== 'claude')→if (!_hostBehaviors(runtime).authorsCanonicalWorkflow). Only the host declares the key, so every other runtime keeps the generic path. -
Route install/uninstall through the public adapter. Replace the direct
installRuntimeArtifacts(...)/uninstallRuntimeArtifacts(...)calls withcreateImperativeAdapter({ runtime }).install({...})/.uninstall({...}). The imperative adapter delegates to the same engine functions, so the output is byte-identical — that is the point: the host is now driven through the interface, not around it. -
Prove parity, both scopes. The differential attribution check (
tests/emitted-attribution.test.cjs, ADR-2719) compares the emitted manifest built from your branch againstnext's recorded state and requires every moved hash to be attributable to a path your PR changed — no fixture to regenerate by hand. Confirm the host's install is unchanged for global and local scopes. Exclude only genuinely volatile / platform-varying files (settings.json,settings.local.json,.gsd-source). -
Guard against regression. Add a
*-imperative-reference.test.cjsasserting the adapter classifies the host correctly, negotiation fails closed on a corrupted descriptor, and — with a source-grep behind an// allow-test-rule:exemption — that noruntime === '<id>'branch remains inbin/install.js.
Another completed worked example: copilot (#2099). Copilot was already installing through the
declarative artifactLayout (not the direct installRuntimeArtifacts calls step 4 describes), so its
migration folded the residual hardcoded branches rather than the whole install path: the .agent.md
destination-suffix rename in src/install-engine.cts (→ hostBehaviors.agentFileExtension), two
uninstall side-effect branches in bin/install.js (→
resolveInstallPlan(runtime).installSurface === 'copilot-instructions', already a live descriptor field
elsewhere in the same file), and two skipSharedHooksInstall gates (→
hostBehaviors.skipSharedHooksInstall: true). A dead legacy agent-converter dispatch arm — unreachable
because copilot was a member of the then-existing _DESCRIPTOR_AGENTS_RUNTIMES allow-list — was deleted
outright rather than re-gated, mirroring step 6's guard:
tests/declarative-reference-copilot.test.cjs source-greps both files for the retired isCopilot reads.
See the copilot section of the reference matrix for the full EoS migration note, including the two
upgrades (multi-event hook bus; negotiated dispatch.background) this PR adds.
_DESCRIPTOR_AGENTS_RUNTIMESno longer exists (#2875). It was an allow-list naming the runtimes whoseagentscame from the descriptor; everything absent from it fell through to an inline_hostBehaviors()dispatch loop inbin/install.js. That loop and the set are both gone — the descriptor is now authoritative foragentson every runtime, so there is no longer an opt-in list to join. Declare anagentsentry underartifactLayoutand it is installed.If your host needs a per-agent transform the descriptor cannot yet express, extend the pipeline rather than reintroducing an inline branch. The three extension points added when the loop was removed are the pattern to follow:
hostBehaviors.agentFrontmatterExtensionsfor injected frontmatter keys, per-agent model-override resolution threaded through the converter's options, and a named converter driven by descriptor data (hermes's branding rewrites are declared incapability.json, not hardcoded). All three exist because the descriptor pipeline lacked one thing: per-agent resolution context (targetDir+agentName).Declaring an
agentsentry also takes effect on the surface path (/gsd-surface --materialize) immediately, not only on install — the two paths are intentionally converged.
Related
- Reference:
docs/reference/host-integration-capability-matrix.md— the per-CLI sourced values. - ADR:
docs/adr/1239-gsd-embeddable-orchestration-engine.md— why the interface exists and the Phase A amendment. - The closed-vocabulary runtime descriptor it extends: ADR-1016.