Fold Copilot's residual runtime-literal branches onto descriptor-driven
hostBehaviors. Several issue premises were inaccurate (verified via research)
and deliberately NOT followed: writesSharedSettings/"legacy exclusion list"
(per-runtime descriptor data, copilot's false is correct); RUNTIME_CONTENT_DISPATCH.copilot
+ installSurface==='copilot-instructions' (already descriptor-driven);
extendedHookEvents (closed Claude/Gemini enum — hooks extended in code instead);
reapply ternary (already folded, kimi #2095).
Real folds (all byte-parity — golden byte-identical for every runtime):
- src/install-engine.cts + src/surface.cts: the two `.agent.md` filename cutovers
(_copyStaged + _syncGsdDir) unified onto hostBehaviors.agentFileExtension via a
new exported agentFileExtensionFor() accessor (kills the two-mechanism divergence).
- src/runtime-artifact-conversion.cts: applyAgentPathRewrites' copilot skip →
hostBehaviors.noPathRewrite:true (antigravity #2096 precedent).
- bin/install.js uninstall: the two isCopilot cleanup branches → installSurface===
'copilot-instructions' gate (symmetric with install-time).
- bin/install.js: `!isCopilot` in the two skipSharedHooksInstall checks →
hostBehaviors.skipSharedHooksInstall:true (copilot has no shared gsd-*.js hooks).
- bin/install.js: three dead legacy inline-agent-loop isCopilot refs removed
(copilot ∈ _DESCRIPTOR_AGENTS_RUNTIMES → unreachable; byte-parity proven by clean
golden + real reachable-runtime install diffs). isCopilot dropped from 4 destructures.
Zero live `runtime==='copilot'`/`isCopilot` branches remain in bin/install.js,
install-engine.cts, surface.cts, or runtime-artifact-conversion.cts (AC2 guard scans all four).
UPGRADE 1 (multi-event hook bus): buildCopilotHookConfig() now emits preToolUse/
postToolUse/userPromptSubmitted/sessionEnd advisory handlers alongside sessionStart
(static inline bash/powershell — deterministic, golden-trackable). Only copilot.json's
gsd-session.json hash changes.
UPGRADE 2 (background dispatch): surfaced via the negotiated contract only —
dispatch.background:true exceeds the declarative-cli baseline and survives negotiation
with no downgrade warning. NO .agent.md frontmatter field (copilot has none). MCP
companion out of scope (AC4 names only 2 upgrades).
Tests: declarative-reference-copilot (adapter/axes/fail-closed + AC2 4-file source-grep
guard) + copilot-upgrades (live 5-event hook wiring; dispatch.background negotiation).
Matrix EoS note + how-to; changeset (Changed). capability-registry regenerated.
Incidental flaky-test RE-ARCHITECTURE (no-defer, maintainer-directed):
tests/opencode-review-reconstruction.property.test.cjs spawned ~600 synchronous
execFileSync('jq') subprocesses (numRuns:200 × 3 fast-check properties, one jq per
generated stream); a single jq freezing on a contended macos-22 CI runner hung the whole
unit-test chunk to its 600s kill (this PR's CI). --test-force-exit can't interrupt a
synchronous execFileSync, so the cure is to stop spawning per case, not just time-bound
it. Re-architected to run the SHIPPED jq program over the whole fast-check corpus in ONE
jq process: each generated stream is one compact-JSON array per line in a temp file,
`jq -c <PROGRAM>` (no -s) applies PROGRAM to each array (`.` == the array, exactly what
production's `jq -rs <file>` sees after slurping) and emits one result per line —
empirically byte-identical to the per-stream form across embedded-newline/empty/quote/
unicode/null-drop cases, and file-input (like production) so there's no stdin pipe to
deadlock on large I/O. ~600 spawns → 6; coverage unchanged (200-case corpus per property,
deterministic seeds) plus explicit boundary/diagnostic example batches. Still property-
tests the real shipped jq (no JS reimplementation). Per-call jq timeout retained as a
belt-and-suspenders bound.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Fold CodeBuddy's residual runtime-literal branches onto descriptor-driven
hostBehaviors (the issue's "zero branches remain" premise was inaccurate — 2
lived):
- Folded the `if (isCodebuddy)` commands/ report duplicate (byte-identical to
the generic reportCommandsDir block above it) onto
hostBehaviors.reportCommandsDir:true, matching cursor's #2089 precedent.
- Deleted the dead `else if (isCodebuddy)` agent-conversion arm (codebuddy ∈
_DESCRIPTOR_AGENTS_RUNTIMES → gated out before the legacy chain, same dead-arm
pattern removed for augment/trae in #2097).
- Removed isCodebuddy from all 4 destructure sites (now comments only, mirroring
the #2096 isAntigravity fold). Zero live isCodebuddy reads remain.
UPGRADE 1 (extended hook bus): populate extendedHookEvents with the full
extended set SubagentStop/Stop/PreCompact/SubagentStart (qwen #2092 / kimi
precedent — codebuddy previously had extendedHookEvents:[] so it got NONE of
these). The generic applySettingsJsonHooks loop wires them, so CodeBuddy's
settings.json now gains all four subagent-lifecycle + stop/compact hooks it
previously lacked, matching Qwen/Kimi coverage. No source change (the
HOOK_EVENT_SURFACES SDK catalog is a locked dict; all 4 events are already in
the validator enum; hookEvents already 'claude'). settings.json is golden-
excluded → no golden change.
UPGRADE 2 (background dispatch): surfaced via the negotiated capability contract
only. CodeBuddy's dispatch.background:true legitimately exceeds the declarative-
cli baseline (false) and survives negotiation with no warning. NO agent-file
frontmatter field is emitted: the CodeBuddy CLI (GSD's install target,
~/.codebuddy/agents/) has NO background-dispatch frontmatter field — background
is a caller-side run_in_background invocation param (verified against
codebuddy.ai/docs/cli/sub-agents); the issue's agentMode/enabledAutoRun are
IDE-only (codebuddy.cn, a different product). Emitting them would be a non-
functional false-green, so it is deliberately not done.
Golden: byte-identical for all 16 runtimes (folds are console/failures-report
only; UPGRADE 1 → golden-excluded settings.json; UPGRADE 2 → no artifact
change). Tests: declarative-reference-codebuddy (adapter/axes/fail-closed
negotiation + AC2 source-grep guard) + codebuddy-upgrades (live install asserts
all 4 extended hooks wired; dispatch.background survives negotiation above
baseline). Matrix + install-on-your-runtime updated; changeset (Changed).
capability-registry regenerated.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Fold augment's runtime-literal conversion branches onto descriptor-driven
hostBehaviors and delete dead code:
- Site A (_applyRuntimeRewrites case 'augment'): the 4 ~/.augment dot-dir
regexes now derive from getDirName('augment') via escapeRegExp (byte-
identical; getDirName('augment')==='.augment') — no runtime literal.
- Site B (applyRuntimeContentRewritesForCommandsInPlace): the
`if (runtime==='augment')` markdown-converter branch now reads
runtime.hostBehaviors.commandBodyConverter and dispatches through a local
COMMAND_BODY_CONVERTERS map (degrade-closed on unknown/absent name).
- Deleted dead `claudeToAugmentTools` map (zero refs; orphaned by ADR-1508
single-sourcing) and the unreachable `else if (isAugment)` agent-conversion
branch (augment ∈ _DESCRIPTOR_AGENTS_RUNTIMES → gated out upstream).
- Incidental orphan cleanup (no-defer): removed the equally-unreachable
`else if (isTrae)` agent-conversion arm left behind by trae's already-merged
migration #2094 (trae ∈ _DESCRIPTOR_AGENTS_RUNTIMES, same upstream gate).
copilot/windsurf/codebuddy arms are removed by their own pending migrations.
UPGRADE 3 (transport:mcp): register the GSD companion MCP server in Augment's
settings.json under mcpServers.gsd (Augment hosts MCP in settings.json, not a
standalone file). mergeGsdMcpServerIntoSettings mutates the in-memory settings
object finishInstall already writes (gated on hostBehaviors.mcpCompanion===
'settings-json'); non-destructive + idempotent; symmetric uninstall removal.
settings.json is golden-excluded, so no golden change. UPGRADE 1 (named/
background dispatch) + UPGRADE 2 (settings-json hook bus, Claude dialect)
were already live in production — this adds tests exercising both.
Golden: byte-identical for all 16 runtimes (folds preserve regex behavior;
MCP lives in golden-excluded settings.json) — verified by a real double-install
tree diff. Tests: declarative-reference-augment (adapter/axes/fail-closed/
undocumented-sub-axes + source-grep guard scoped to conversion-logic branches)
+ augment-upgrades (dispatch negotiation, hook-bus live install, MCP add/
idempotent/preserve/uninstall). Matrix + connect-gsd-mcp-server + a stale
install-on-your-runtime hook-ownership claim corrected; changeset (Changed).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Fold all runtime==='kimi'/isKimi logic branches into descriptor-driven
hostBehaviors (localInstallDeferred, verificationStyle, agentManifestStyle,
reapplyCommand, doneBannerStyle) + add 'kimi' to _DESCRIPTOR_AGENTS_RUNTIMES.
Kimi's skills/kimi-agents dispatch was already descriptor-driven (converter-by-
name + kimi-agents kind). Zero isKimi/runtime==='kimi' branches remain.
UPGRADE 1 (native hook bus): new hooksSurface 'kimi-hooks-toml' + a marker-
delimited config.toml [[hooks]] emitter (buildKimiHooksTomlBlock/writeKimiHooksToml
in runtime-hooks-surface.cts; resolveKimiHooksTomlDir in runtime-homes.cts).
GSD's lifecycle hooks now wire into Kimi's native ~/.kimi/config.toml (Context7-
confirmed path) at SessionStart/PreToolUse/Stop/PreCompact/SubagentStart/
SubagentStop — kimi becomes a hooks/ consumer (the 3 && !isKimi exclusion guards
removed). config.toml holds absolute install paths so it's golden-excluded via
an exact relative-path (.kimi/config.toml), not a basename (which would blind
Codex's config.toml). New hooksSurface value added to the closed enum in
capability-validator + runtime-config-adapter-registry.
UPGRADE 2 (background dispatch): flip dispatch.backgroundDispatch true (Kimi's
Agent tool takes run_in_background; root agent already gets the Agent tool), so
negotiation no longer flattens dispatch. subagentToolkit stays 'undocumented'
per AC (coder/explore/plan have distinct tool policies).
MCP transport explicitly deferred (no installer-driven MCP for any runtime).
Golden: only kimi.json changes (hooks/ scripts now installed); all 15 others +
claude-local byte-identical (kilo/zcode keep their own exclusions). Tests:
kimi-imperative-reference (adapter/axes/fail-closed/hostBehaviors + source-grep
guard) + kimi-upgrades (config.toml [[hooks]] SessionStart + marker idempotency
+ backgroundDispatch negotiation). CONTEXT.md glossary + matrix + how-to updated;
changeset (Added).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Fold trae logic branches into descriptor-driven reads: skipSharedHooksInstall
gates (dropped && !isTrae), and the case 'trae' path-rewrite arm now computes
the self-alias from the descriptor-driven dirName (.trae). Dead isTrae bindings
removed from uninstall/writeManifest/finishInstall. trae's skills dispatch was
already descriptor-driven (converter-by-name). RUNTIME_CONTENT_DISPATCH.trae is
left as a runtime-keyed table registration (its regex/callback rewrites can't be
a byte-identical descriptor map — matches cursor/windsurf/cline). trae stays in
RUNTIME_FLAG_IDS: isTrae still gates the agents-converter selection (agents out
of scope; removal gated on the cross-runtime agents-dispatch migration).
Byte-identical golden parity for all 16 runtimes.
UPGRADE: SOLO stage/trigger metadata — emitted Trae SKILL.md now carries
stage: workflow (descriptor-gated via hostBehaviors.soloStageMetadata) so
Trae's SOLO Agent can auto-invoke GSD skills at the corresponding stage. Field
shape is best-effort/inferred (Trae publishes no formal schema). trae.json
golden regenerated.
Tests: trae-imperative-reference (adapter/axes/fail-closed shouldFlattenDispatch
+ no runtime==='trae' source-grep, isTrae exempted for agents) + trae-upgrades
(stage: workflow on installed SKILL.md, descriptor-gated). Matrix note +
changeset added.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Fold remaining isKilo logic branches into descriptor-driven reads:
finishPermissionWriter (uninstall cleanup), skipSharedHooksInstall (hooks
copy), and a skills converter-name registry (the artifactLayout.converter
field is now load-bearing, not decorative). frontmatterDialect stays the
documented dispatch key for frontmatter (no descriptor field for it). Dead
isKilo destructure bindings removed. Byte-identical golden parity for all 16
runtimes (opencode, which shares kilo's combined-family path, verified clean).
UPGRADE 1 (hook bus): install .kilo/plugins/gsd-core.js native plugin +
extensionEvents:"kilo" + EXTENSION_EVENT_SURFACES.kilo (OpenCode-fork bus).
UPGRADE 2 (active model): populate runtimeTierDefaults.kilo + thread
modelOverride through convertClaudeToKiloFrontmatter — model no longer stripped
from agents. UPGRADE 3 (MCP): document the gsd-core MCP companion under kilo's
mcp config key. UPGRADE 4 (named dispatch): agents/*.md mode:subagent roster is
the Task-tool dispatch surface (tested); subagentToolkit stays 'undocumented'
per AC so dispatch degrades to 'degraded' by design.
Model-catalog single-source edit ripples the shared model-catalog.json hash
into all 16 golden fixtures (expected). Inline defect fixes (no-defer): stale-
bake-guard resolveAgentDir 'agent'->'agents' (was a silent no-op for opencode/
codex), hardcoded 'Removed OpenCode plugin' uninstall log -> generic, and the
connect-gsd-mcp-server.md OpenCode mcpServers->mcp doc error.
Tests: kilo-imperative-reference (adapter/axes/fail-closed/degradation/
hostBehaviors + widened isKilo source-grep across 4 modules) + kilo-upgrades
(plugin parity+load, model-override converter, agents dispatch surface, MCP
doc). Matrix + how-to + config docs updated; changeset added.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
trek-e's re-review asked to either mechanize the Step 9.5 element
extraction or add an inline note justifying why it stays manual, so a
future maintainer doesn't read it as an oversight.
Verified the premise: the requirement-side edge-probe path (spec-phase
Step 5.5) it mirrors is ALSO a hand-populated heredoc + fail-loud
<replace:> placeholder guard — not a mechanical parse. The UI probe
mirrors that idiom verbatim. Mechanizing would be worse: a UI-SPEC has no
single machine-parseable "elements" column (surfaces are spread across the
design-token tables, Copywriting, and researcher-named prose), so a
regex/table parse would fail-OPEN (miss a prose-named surface, or feed a
design-token row as a bogus element).
Expanded the inline comment to state the parity + the fail-open rationale
explicitly. No logic change; regenerated the workflow size baseline and
the 16 golden-install-parity fixtures for the +847 B comment.
Refs #1867
Claude-Session: https://claude.ai/code/session_017vYn26e3nkDNxcpty1ciPJ
Record resolved UI-state considerations in the UI-SPEC, backward-compatibly
(Phase 2, WIRE-02). templates/UI-SPEC.md gains a '## UI Considerations' section
(analog of SPEC '## Edge Coverage') after '## Copywriting Contract' — a
| Category | Element(s) | Status | Resolution / Reason | table with
covered/backstop/unresolved rows in the locked probe-core projectTruths format
the shipped plan-phase lift (plan-phase.md:921) reads. Empty/error COPY stays in
Copywriting; this section covers shape-rooted STATE and references those rows
(de-dup).
Tests: docs-fixtures parsed-heading assertion (UI Considerations present +
distinct from Copywriting Contract; allow-test-rule, parsed structure); typed
backward-compat (projectTruths(undefined/[])===[], old UI-SPEC still plans —
Hyrum), format-match, and idempotency (proposeElements determinism). The
template-structure test was the RED driver.
Install-parity cascade (INVENTORY-MANIFEST + 16 golden fixtures +
agent-size-baseline) deferred to Phase 3 SHIP-01, as planned.
Claude-Session: https://claude.ai/code/session_01BKt4hgNZwXSeJYJtYAQUSS
Add the live ui-phase producer path for the UI-consideration probe (Phase 2,
WIRE-01). Two small exports on the Phase-1 adapter — proposeElements (the
propose-then-confirm view of detected kinds + applicable categories) and
autoResolve (the deterministic --auto floor that never dismisses and never
auto-backstops an unclassified item, #1110) — plus a post-verification
'## 9.5 UI-Consideration Probe' step in ui-phase.md mirroring spec-phase 5.5's
RUNTIME_DIR shim + fatal-invoke/malformed-report/zero-applicable fail-closed
guards, propose-then-confirm (the partial-cue recall mitigation), and the
'## UI Considerations' write-back in the shipped plan-phase.md:921 lift format.
autoResolve is the CODE floor; the covered-upgrade stays workflow prose (the
two-layer --auto). Un-upgraded backstops route to insufficient_spec ->
human_needed at verify, never a silent pass (#1154).
Tests: +8 typed (proposeElements shape/determinism, autoResolve never-dismiss,
partial-cue strict-subset) — structured-value only. ui-phase.md size baseline
ratcheted 15477->24447 (under DEFAULT cap). The plan-phase.md PRE_PHASE6 ceiling
stays RED pending #1852 (unchanged from Phase 1).
Claude-Session: https://claude.ai/code/session_01BKt4hgNZwXSeJYJtYAQUSS
Separate *-UI-SPEC.md glob (edge-coverage glob exclusion at :746 left intact
- Hyrum/D-08); a terse lift bullet reusing the ## Edge Coverage rule verbatim
(covered -> truths string, backstop -> flat scalar {statement, verification:
backstop}, unresolved -> assumption; no new verb - ADR-550 #1278/#1154); and a
no-silent-drop checklist line. Lift logic lives ONLY in the plan-phase
workflow, never gsd-planner.md (D-10, agent-size cap).
NOTE: this grows plan-phase.md +862B, over the #1168 PRE_PHASE6 ceiling (94519)
by ~802B, so the workflow-size gates are RED until #1852's lazy-split lands the
-21KB headroom. Deliberately does NOT raise the ceiling constant (would collide
with #1820/#1835's in-flight raise). #1867 sequences after #1852; rebase +
regenerate the size baseline then.
Reference doc mirrors edge-probe.md structure but links rather than re-argues;
states the MIXED-axis boundary (closed compiled shape-rooted subset here; open
UX subset - real-time/offline, a11y depth, i18n/RTL - prose-owned in
domain-probes.md). Docs-parity test pins doc taxonomy ids == code UI_TAXONOMY
ids and asserts disjointness from domain-probes.md topics (ADR-456
runtime-contract exemption, see #1867).
Update runtimeTierDefaults.codex and providerPresets.openai in
model-catalog.json to the GPT-5.6 family (gpt-5.6-sol/terra/luna),
advancing from the superseded GPT-5.4/5.5 generation.
Model IDs verified against OpenAI developer API docs:
- gpt-5.6-sol: flagship, /, reasoning xhigh
- gpt-5.6-terra: balanced, .50/, reasoning medium
- gpt-5.6-luna: fast/cheap, /, reasoning medium
Tier mapping is 1:1 (Sol↔flagship, Terra↔balanced, Luna↔fast),
so profile semantics are unchanged — only the underlying IDs advance.
Updates: catalog JSON, test assertions (catalog defaults), docs
(CONFIGURATION.md + zh-CN/pt-BR translations, workflow settings),
and changeset.
Closes#2122
Review caught that the prior commit bounded only the paren tag clause and left
the SIBLING bracket-prefix `(?:\[[^\]]+\]\s*)?` (same host regexes, before Phase)
UNBOUNDED — the identical quadratic reachable via a `[...]` run (measured ~16s at
1.7MB). Bound `[^\]]+`/`[^\]]*` -> {1,200}/{0,200} across all 19 phase/milestone
heading prefixes. Comprehensive re-measurement now shows EVERY vector linear
(bracket/paren/id/name/milestone all ~2-44ms at 2.45MB; bracket scaling
2k->2ms, 4k->5ms, 8k->10ms). Also: update the #1729 literal-mirror parity test
off its stale unbounded constant, and add limit-1 (199) boundary coverage.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The previous registry was generated from a stale main-repo state that was
missing cline's hostBehaviors (merged in #2090). CI's lint:generated-sync
detected the staleness. Regenerated from clean origin/next + hermes changes.
Drive Codex install/uninstall through the descriptor-driven Host-Integration
Interface (declarative embedding adapter → engine surface dispatch) and fold
every positive `runtime === 'codex'` / `isCodex` projection into descriptor-driven
`runtime.hostBehaviors`. Install/uninstall output stays byte-parity-gated
(tests/fixtures/golden-install-parity/codex.json); no other runtime changes.
Three Context7-verified upgrades, each with a test on the user-reachable surface:
- Skill root → canonical $HOME/.agents/skills via a skills-kind `home` override,
with pre-move migration cleanup (stale ~/.codex/skills/gsd-* removed on install
and uninstall; user content preserved). Fixes getGlobalSkillsBase, writeManifest,
and the skill-manifest inventory to honor the override so --skills-root /
sync-skills / the manifest report the real location.
- Six new hooks.json lifecycle events (PreToolUse, PermissionRequest, PreCompact,
PostCompact, SubagentStop, UserPromptSubmit) shared by install + uninstall;
extendedHookEvents reconciled [] -> the schema-valid wired subset.
- Explicit `[agents] max_depth = 1` in the managed config.toml block, pinning the
negotiated dispatch.maxDepth:1 axis. validateCodexConfigSchema now permits a
known-scalar-only bare `[agents]` AgentsToml table (still rejects [[agents]] and
unknown-key break-forms, #2760); mergeCodexConfig preserves the user's own
AgentsToml scalars (max_threads etc.) instead of dropping them.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Code review (HIGH): a hardcoded 'timeout 600 agy' fails with rc 127 on stock
macOS (no GNU timeout/gtimeout), silently losing the agy reviewer. Probe for
'timeout'/'gtimeout' via command -v and fall back to agy's native --print-timeout
alone when neither exists (mirrors scripts/base64-scan.sh). External cap (600s)
stays >= --print-timeout (540s) so it only backstops a pre-session stall. Factor
the prompt into _AGY_PROMPT to avoid duplicating the long -p string across both
branches. Update the agy + #687 tests to assert the probe + bound + fallback,
regen the 17 goldens + size baseline, refresh the maintainer-note version stamp
to 1.0.16.
Route OpenCode (and its Kilo sibling) through the public Host-Integration Interface and
land two Context7-verified capability upgrades. Byte-identical install output for all 16
runtimes (golden parity asserted).
Through the interface (AC2):
- OpenCode/Kilo's bespoke commands+skills+plugin install (the inline
`else if (isOpencode || isKilo)` block) moves into the engine
(installOpencodeFamilyCommands/Artifacts in src/install-engine.cts), dispatched by
installRuntimeArtifacts when the descriptor declares hostBehaviors.combinedFamilyInstall.
opencode/kilo now flow CLI -> _runtimeAdapter -> installRuntimeArtifacts like the skills
runtimes. _isSkillsRuntime no longer excludes them; the bespoke block + dead
copyFlattenedCommands are removed.
- Every hardcoded `runtime === 'opencode'`/`isOpencode` branch is folded into
descriptor-driven runtime.hostBehaviors. ZERO `runtime === 'opencode'`/`'kilo'`
string-equality remain in bin/install.js / install-engine.cts / runtime-artifact-conversion.cts.
Upgrades (AC4):
- Background dispatch: OpenCode shipped experimental background subagents in v1.15 and
made them default-on in v1.17 -> dispatch.background/backgroundDispatch flip to true;
shouldFlattenDispatch(opencode) now returns false (behavioral change; type: Changed).
- Expanded event surface: the OpenCode plugin subscribes permission.asked/replied +
session.error.
Tests: opencode-imperative-reference (adapter/profile, shouldFlattenDispatch pin,
fail-closed negotiate, hostBehaviors, AC2 source-guard) + extended plugin surface test.
Docs: capability matrix v1.15/v1.17 citations. Changeset (Changed). gitignore .memdb//.memtrace/.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Three failure modes on agy 1.0.16, all fixed by mirroring the Cursor block's
invocation discipline:
* file-reference prompt instead of inline "$(cat)" — a large review prompt
overflowed the exec arg list (rc 126).
* external 'timeout 600' wrapper — --print-timeout cannot fire before agy
creates a session, so a pre-session stall hung unbounded.
* --model from review.models.agy when set — escape hatch for a pinned model
that 404s (exit 0, empty stdout + transcript).
* stdin </dev/null so agy never blocks on a tty.
Also enrich the Step 3 empty-output stub to grep agy cli.log for a
model-availability diagnostic, and correct the stale 'no --model flag' note
plus the 'review.models.agy reserved for future' comment (the config key was
already read but never passed through).
Fold Claude Code's install/uninstall onto the Embeddable Orchestration System
(ADR-1239 Phase D). claude is GSD's tier-1 reference host, but its install path
was still driven by 13 hardcoded `runtime === 'claude'` string-equality branches
scattered across bin/install.js rather than the public Host-Integration Interface.
- Route install()/uninstall() through `createImperativeAdapter({runtime})` — the
adapter delegates to the SAME installRuntimeArtifacts/uninstallRuntimeArtifacts
engine calls, so output is byte-identical (proven pre/post, both scopes).
- Replace all 13 `runtime === 'claude'` / `runtime !== 'claude'` branches with
descriptor-driven `runtime.hostBehaviors` lookups on capabilities/claude/
capability.json (attributionSource, authorsCanonicalWorkflow, localInstallStyle,
permissionsSchema, settingsFileByScope, sourceMarkerFile, agentFrontmatterExtensions,
ownsClaudePaths, nativeModelAliases, skillsGlobalOnboarding). Behavior is
identical; the brittle string-equality coupling (the add-a-host tax) is gone.
- Single-source the scattered literal 'claude' defaults/rosters behind DEFAULT_RUNTIME.
- Extend golden-install-parity to assert the claude LOCAL legacy layout is
byte-identical too (AC1 "both scopes"); exclude the platform-varying
settings.local.json (same reason settings.json is excluded).
- New tests/claude-imperative-reference.test.cjs: adapter kind, programmatic-cli
profile, fail-closed negotiation on a corrupted/partial descriptor, and an AC2
source guard that no `runtime === 'claude'` branch remains.
No user-visible install-output change (internal architecture only).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
model_overrides / models.<phaseType> were silently inert for gsd-assumptions-analyzer,
gsd-code-reviewer, and gsd-code-fixer on Claude Code: resolveModelInternal honors them,
but the workflows spawned these agents with no model= param, so the resolved value
never reached the Agent tool and the agents inherited the session model — no warning.
Fix — thread each agent's resolved model at every spawn site (the established
plan-phase pattern; the architecture-consistent Claude mechanism, since 13 other
agents already thread their model):
- discuss-phase-assumptions.md: `resolve-model gsd-assumptions-analyzer --raw`
→ ANALYZER_MODEL, threaded.
- code-review.md + code-review-fix.md (re-review): `resolve-model gsd-code-reviewer --raw`
→ REVIEWER_MODEL, threaded.
- code-review-fix.md (both fixer spawns): `resolve-model gsd-code-fixer --raw`
→ FIXER_MODEL, threaded (same silently-inert bug, same file — folded in per review).
- quick.md review step: was reusing `{executor_model}` for gsd-code-reviewer (so the
reviewer's own override was ignored); init.quick now resolves `reviewer_model`
(gsd-code-reviewer) and the spawn threads it.
resolve-model --raw returns the bare model string (resolve-execution --raw would
return effort — wrong). The resolver maps these agents to phaseType discuss /
verification / execution, so models.<phaseType> apply too.
Scope: the three agents reachable from the two issue-named workflows + quick.md. The
wider systemic class (other agents in UNTOUCHED workflows with the same pattern) stays
documented on the issue for a maintainer-scoped structural decision (thread-at-source
vs embed-at-install like #2256), not widened here.
Docs: the stale "discuss — reserved, no subagent today" model-profile tables now list
gsd-assumptions-analyzer and the verification row includes gsd-code-reviewer, across
the English docs, the shipped gsd-core/references/model-profiles.md reference, and the
ja-JP / zh-CN / ko-KR / pt-BR locale mirrors.
Tests:
- tests/model-resolver.test.cjs: #2072 acceptance — model_overrides and
models.discuss/verification/execution resolve for all three agents.
- tests/model-routing-spawn-threading.test.cjs: every spawn of the three agents threads
a resolved model (fails pre-fix); a header-precise parity guard fails the suite if a
new un-threaded spawn of any of them regresses.
All 16 golden-install-parity fixtures + the workflow size baseline regenerated for the
changed shipped files (4 workflows + the reference doc); bin/lib is excluded from parity.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Full API Coverage by Default — Opt Out, Never Opt In. A phase that integrates
an external API/SDK/service can no longer seal without a decided coverage matrix.
- src/api-coverage.cts: deterministic detector (compound verb+noun signal +
<Service> API/SDK surface; stopword-guarded; strips fenced code) + matrix
parse/validate/render with field-length caps.
- check api-coverage.verify-pre: blocking seal-time gate; phase arg resolved as
a token under .planning/phases/ only (traversal-neutralized); validates
COVERAGE.md or blocks iff a strong integration signal is detected and no
matrix exists; fail-closed when phases tree exists but phase unresolvable.
- capabilities/ai-integration: workflow.api_coverage_gate config key (default
true), plan:pre contribution, blocking verify:pre gate. Data-driven.
- gsd-core/workflows/verify-work.md: generic verify:pre gate dispatch.
- Tests: detector FP/FN + matrix validation + fast-check bijection; gate e2e.
Code+security review findings fixed (stopword FP, scope containment, pipe/cap
rejection, prompt-injection message hygiene).
- Regenerated registry/matrix/loop-host-contract/goldens/baseline + docs.
Closes#1562
A GSD verification gate resolves a project's test command and runs it. vitest
defaults to WATCH mode in an interactive TTY — exactly where a user runs
`gsd-execute-phase` — so a resolved `npm test`/`pnpm test` backed by vitest never
exited and the orchestrator waited indefinitely. Recovery needed the user to
manually prompt "something blocking?".
Fix — one shared helper + a bounded, surfacing timeout on the test-command gates:
- New pure module src/normalize-test-command.cts + `gsd-tools query
normalize-test-command` verb: rewrites a resolved command to a best-effort
one-shot form (direct vitest → `vitest run`; jest `--watch` → `--watchAll=false`;
a package-manager `test` script whose package.json runner is watch-vitest →
`CI=true` prefix; handles `--dir`; already-one-shot commands unchanged —
never double-flagged). Named `normalize-test-command` (not `test-*`) so the
file does not match node --test's default `test-*` discovery glob.
- The three gates that HUNG or silently-continued route through that ONE helper
and bound execution with `timeout $(config-get workflow.test_gate_timeout)`
(new config key, default 600s): the regression gate (extracted to
execute-phase/steps/regression-gate.md since execute-phase.md is size-frozen —
it shrank 93528→93132; ABORTS on exit 124), the post-merge gate, and the
audit-fix gate (previously an UNBOUNDED `eval`). All name watch/dev mode on 124.
- verify-phase's gate was ALREADY bounded (a fixed `timeout 300`, not a hang), so
it only gains the normalizer (so a watch runner exits fast) + a watch-mode hint
on 124, staying under its frozen 40960-byte tier cap.
Security hardening (review): the normalizer only rewrites a runner named as a
standalone command TOKEN (so `run-vitest.js`/`make test-vitest`/paths are never
mangled), is length-capped and uses only linear-time split-based scanning (no
super-linear backtracking on an adversarial `workflow.test_command`), and reads
package.json only when it is a regular file (never blocks on a FIFO via `--dir`).
Config key `workflow.test_gate_timeout` (seconds, default 600) registered in the
schema manifest + templates/config.json + docs/CONFIGURATION.md (mirrors
workflow.cross_ai_timeout). New module registered in .gitignore, eslint ignores,
inventory manifest/index. All 16 golden-install-parity fixtures + workflow size
baseline regenerated for the changed shipped files; bin/lib is excluded from the
parity manifest.
Tests: tests/normalize-test-command.test.cjs (normalizer units incl. security
hardening) and tests/test-gate-watch-mode.test.cjs (the three core gates route
through the shared helper + configured timeout + exit-124 watch-mode hint;
verify-phase asserted as normalize-only/already-bounded).
tests/execute-phase-active-flags.test.cjs repointed at the extracted step;
tests/planner-language-regression.test.cjs allowlist comment updated.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
A skills-only role:feature third-party capability installed 'active' but its skills never reached the runtime surface, capability enable/set rejected it as 'unknown capability', and capability list disagreed with capability state. Three defects, fix shape 1b (teach resolveSurface, no on-disk linking):
D1 (materialization): resolveSurface built the surfaced skills Set only from the on-disk manifest; third-party cap skills live at ~/.gsd/capabilities/<id>/skills/ and never entered the Set -> surfaced:false. Fix: union registry.capabilityClusters values into the Set in the full-profile branch (idempotent for first-party, additive for third-party; prototype-pollution guarded).
D2 (enable/set unknown): setCapabilityState validated the capId against the static first-party registry instead of the composed overlay-aware registry. Fix: validate against loadRegistry({includeInstalled}) like capability-state.cts does.
D3 (list vs state): capability list derived status purely from ledger existence, never surface composition. Fix: add a 'surfaced' field to list rows sourced from the same resolver capability state uses, so list and state agree.
Regression coverage: tests/issue-2045-third-party-skills-surface.test.cjs asserts all five acceptance criteria (resolveSurface union, state surfaced:true, enable/set not-unknown, list/state agreement, first-party + unknown-id regression). gsd-test: 23916/23916 green (linux-node22 + linux-node24).
resolveCapabilityRuntimeState derived the config dir from resolveRuntime(cwd)
(GSD_RUNTIME -> config.runtime -> 'claude') when no --config-dir was passed,
so a repo with persisted runtime:'codex' resolved the config dir to ~/.codex
where the Claude skill isn't installed -> surfaced:false / hooks silently
no-op when the operator drove from Claude Code. capability state and loop
render-hooks parsed only --config-dir, never --runtime, so there was no way
to assert the actually-active runtime.
Add a runtimeOverride param to resolveCapabilityRuntimeState (canonicalized
via runtime-name-policy so aliases like codex-app work); when present it
short-circuits the persisted-runtime fallback and resolves getGlobalConfigDir
for the explicit runtime. Thread --runtime through cmdCapabilityState and
cmdLoopRenderHooks, and parse it in gsd-tools.cjs for both commands (dual
--runtime X / --runtime=X form, mirroring --config-dir and the existing
capability-set --runtime precedent). Help text updated.
Without the override, behavior is byte-identical to today (regression-guarded).
- compareSemver: implement full SemVer 2.0.0 §11 pre-release identifier
comparison (two pre-releases of the same triple now order correctly; was 0).
- capability description + fragment: scope the plan-checker/verifier claim
(this capability delivers the parallel-execution backend; those gates remain
inline until separately wired). Correct the 'each wave is one barrier' prose
(a wave splits into multiple sequential parallel() barriers on files_modified
overlap). Frame detect-backend CLI as a simulation harness; the pure function
with the live host descriptor is the real detection seam.
- partitionStages docstring: 'near-minimal via greedy first-fit' (not 'fewest');
document empty-files_modified behavior.
Claude Code's isolation="worktree" forks new worktrees from origin/HEAD, not
the live local HEAD. When prior local commits (e.g. an earlier quick task in
the same session, or this task's own Step 5.6 pre-dispatch plan commit)
advance local HEAD without an intervening push, origin/HEAD stays pinned to a
stale ancestor and the executor's worktree_branch_check guard halts with a
base-mismatch fatal that can be many commits behind, not just one.
Port the worktree.base-check auto-degrade pattern already used by
execute-phase (#683/#1369) into quick.md's single-dispatch path, run
immediately before EXPECTED_BASE is captured in Step 6.
- Teach applySurface to build agentCtx (pathPrefix + attribution) and pass it
to kind.stage() for agents kind, mirroring createRuntimeArtifactInstallPlan
(ADR-1235 §1). Surface-path agents now receive path-rewrite + attribution +
converter + normalize, matching install output byte-for-byte.
- Pass skills:'*' sentinel for agents staging when no surface state modifications
exist, so ALL agents are staged (not just those referenced by _calls_agents_).
- Declare converted agents kind in copilot and antigravity capability.json;
add to _DESCRIPTOR_AGENTS_RUNTIMES in bin/install.js.
- Handle copilot .agent.md filename rename in both _copyStaged (install path)
and _syncGsdDir (surface path).
- Ship golden-parity harness (ADR-1235 §0): tests/issue-1575-agent-descriptor-
parity.test.cjs asserts applySurface output is byte-identical to
installRuntimeArtifacts for all 7 descriptor-driven runtimes, plus stale-
cleanup convergence and prune data-loss coverage.
- Update ADR-1235 with cutover progress.
Cline remains deferred (rules-only local branch + local/global complication).
Resolve the zcode test cascade exposed by gsd-test:
- model-catalog.json: add zcode runtimeTierDefaults (null tiers, matching the
other profile-marker-only runtimes) so KNOWN_RUNTIMES stays parity with allRuntimes.
- runtime-name-policy: add zcode to GLOBAL_CONFIG_HOME_FRAGMENTS (~/.zcode) so
getGlobalConfigHomeFragment stops falling through to the .claude default.
- installer-migration-install: add the zcode fresh-install contract (flat-skills,
no settings, no package.json — same shape as trae).
- golden-install-parity: capture the zcode fixture via a standalone generator
script (not node --test — the gate stays gsd-test).
- capability-matrix.md: regenerate so zcode appears as a first-party row.
Fluidify the remaining count-pinned guards so adding a runtime no longer trips a
hand-pinned snapshot: gemini-runtime-removed (flag count derives from the
registry), non-claude-runtimes-registry-derivation (golden list derived from the
registry).
Add ZCode as a first-party runtime via a declarative capability descriptor
(capabilities/zcode/capability.json) with zero hardcoded runtime === 'zcode'
branches — exercising the de-hardcoded, data-driven runtime path that 1.7.0
(ADR-1016 / ADR-1239) enables.
Descriptor (all axes sourced verbatim from zcode.z.ai docs):
- configHome ~/.zcode; nested skills + flat commands/agents; profile-marker install
- Claude-shaped skill format reuses convertClaudeCommandToClaudeSkill (no new converter)
- hostIntegration: declarative / slash-file / mcp / electron; dispatch background=false
(foreground-only per docs); nested+maxDepth undocumented; passive model mode
Installer registration (data, not branches): --zcode flag, allRuntimes, runtimeMap,
interactive menu, --all list. getGlobalConfigDir + resolveRuntimeArtifactLayout +
ALLOWED_CONFIG_RUNTIMES + resolveInstallPlan all derive from the descriptor.
Revamped the brittle per-runtime golden-master tests to be count-agnostic,
descriptor-derived property tests (1.7.0 makes runtimes pluggable data, so pinning
frozen '15 runtime' snapshots is the wrong invariant): getdirname, label-policy,
config-adapter-registry (intent + install-plan golden master), capability-registry,
host-integration-descriptors (counts derive from curated maps). Adding a runtime
descriptor now extends coverage with zero edits to those suites.
Welcome banner, --zcode help, supported-runtimes how-to, and the host-integration
capability matrix (every axis cited) updated.
