* test(#2871): failing-first suite for trigger-surface resolution 23 tests over the 50-test-matrix rows. RED by construction: resolveTriggerSurface and DEFAULT_TRIGGER_PRECEDENCE do not exist yet, and the validator silently ignores triggerPrecedence today. Written in the per-runtime describe idiom the other four runtime-artifact-layout suites use, not a table. The rows that carry the weight: windsurf must NOT report a shadow it does not have, since its global scope emits only agents and agents are not trigger-bearing; agents and kimi-agents must be absent from the output for every runtime; and reordering a runtime's triggerPrecedence must flip the winner, which is the only assertion that proves the axis is read rather than decorative. Stems are injected, never scanned, so the surface is assertable with no filesystem. * feat(#2871): resolve triggers and host precedence, not just placement resolveTriggerSurface(runtime, scopes) returns every /gsd-<name> trigger a runtime emits, with the scope and kind that produced it, whether the host registers it directly or only through a router, and which artifact shadows it. resolveRuntimeArtifactLayout is untouched -- its 7 callers need placement only and the issue requires them unchanged. AGENTS ARE NOT TRIGGER-BEARING, and ADR-2866 said they were. The host-integration matrix models command and dispatch as separate interface points: an agent is invoked through the Agent tool's subagent_type, not by typing a slash trigger, and _copyStaged never applies the kind prefix to an agents entry. So agents and kimi-agents are excluded from the surface entirely, and this commit amends ADR-2866 with a dated correction. #2218's conclusion is unchanged -- the collision is strictly commands-vs-skills, and claude's local /gsd-* trigger surface is still fully shadowed -- but the ADR implied the local agents surface was lost too, and it is not. That correction is what makes windsurf come out right. Its global scope emits only agents, so it has no global trigger and its local commands are unshadowed. Model agents as trigger-bearing and windsurf falsely reports a full shadow. The triggerPrecedence axis lands on all 19 descriptors as an ordered kind list, one value with one owner, rather than a numeric rank spread across N kind entries with nothing keeping them consistent. Validation uses a required-with-default shape that has no precedent in this validator -- every existing axis is hard-required -- so a third-party capability.json omitting the field still validates, which is what ADR-894's additive-only contract promises. Winner resolution reads Phase 1's scope rank first, then the kind ordering. A test reorders the axis and asserts the winner flips, since an axis that is added, validated and never consulted would pass every other assertion. shadowedBy ships unread. Phase 4 (#2873) is its first consumer, per this issue's out-of-scope note. Verified via the remote runner. * fix(#2871): single-source namespacedByDir and close two test gaps Four findings from the isolated adversarial review. The namespacedByDir rule had reached three copies -- install-engine, surface, and the new trigger resolver -- one of which carried a hand-written keep-in-sync comment and no assertion. That is this repo's generative-fix-divergence class. Extracted to one exported predicate all three now call. Verified by diverging one copy deliberately: the existing #816 parity test failed, and passes again on revert. The omission test was vacuous. Row 16 asserted that a descriptor without triggerPrecedence still validates, but built its fixture from claude's shipped descriptor -- which this PR had just added the axis to. It now clones and deletes the key, following the shippedDescriptorWithout pattern, and asserts both that validation passes and that the resolver still picks the right winner from the default. The second half is what makes it prove anything. resolveTriggerSurface silently dropped an unrecognized scope while every sibling in this epic throws. Two phases of one epic should not disagree about whether an invalid scope is an error, so it now rejects through the same shared validator; an empty scope list still returns empty rather than throwing. The ADR amendment had been spliced into the middle of the References list, orphaning its last bullet. Moved to the top, after the header block, which is where ADR-3660 and ADR-1016 both put dated amendments. No lint checks markdown structure, so this was green while malformed. * fix(#2871): single-source the command filename composition too The earlier fix shared the namespacedByDir boolean but left the filename composition around it written twice -- once in _copyStaged as what actually gets written, once in resolveTriggerSurface as what gets predicted. The predictor could go stale silently. One exported helper now composes it for both. The entry.name asymmetry that looked like it would block extraction does not: entry.name is filtered to end in .md and stem is entry.name minus those three characters, so the two branches are the same string by construction. Divergence proven to fail: injecting a marker into the helper broke the trigger-surface suite; reverting restored 25/25. The four sibling layout suites hold at 227 unchanged. * docs(#2871): correct the ADR timing notes that this phase makes stale The Amended by back-links on ADR-3660 and ADR-1016 were written in Phase 0, when the widenings they describe had not shipped. Each carried a forward-looking clause -- "the module changes at Phase 2, not before, until then this module resolves placement only" -- which becomes false the moment this PR merges. ADR-2866's own Amends header and its reciprocal-notes section carried the same tense. All four now describe what shipped. This is a tense and status correction on Accepted ADRs, not a change to any decision. Worth stating because it is the failure mode this epic keeps meeting: gen-adr-index.cjs tracks only Supersedes and Subsumes, so nothing in CI would have caught either the missing back-link in Phase 0 or these stale clauses now. They stay correct only because someone checks. * chore(#2871): backfill changeset PR number --------- Co-authored-by: sim <sim@local>
18 KiB
Runtime Artifact Layout Module owns per-runtime artifact placement
- Status: Accepted
- Date: 2026-05-17
- Issue: #3660
- Implementation: #3663 (Phase 1), feat/3663-runtime-artifact-layout-module-phase-1-m
- Subsumed by: ADR-1239 (GSD as an Embeddable Orchestration Engine) — read it first; see the amendment below
- Amended by: ADR-2866 (Install-surface resolution) — this module widens from placement to placement + trigger resolution. The
Layoutreturned here models where a file goes but not the/gsd-<name>trigger it occupies, so nothing in the tree can express the cross-scope collision in #2218. ADR-2866 adds a projection returning the resolved trigger, its kind and scope, its destination, and whether another install shadows it. This ADR's placement decision is unchanged and still in force —resolveRuntimeArtifactLayoutstays for callers that only need placement, the per-runtime table remains the single owner of placement knowledge, and legacy-layout migrations stay in ADR-0008. Timing: Phase 2 (#2871) of epic #2866 has landed — the module now resolves placement and triggers via the newresolveTriggerSurfacefunction;resolveRuntimeArtifactLayoutis unchanged for callers that need placement only.
Amendment (2026-07-16): subsumed by ADR-1239 (EoS)
ADR-1239 — GSD as an Embeddable Orchestration Engine (EoS), Accepted — subsumes this ADR as an adapter: per-runtime artifact placement becomes one negotiated surface of the Host-Integration Interface rather than the outermost seam at which GSD meets a host.
This ADR is not superseded and its status is unchanged. The artifact-layout seam is live and load-bearing; it is now a component of the EoS frame.
Read ADR-1239 first.
Recorded because ADR-1239 declared this subsumption while this file recorded nothing.
Context
The Runtime Surface Module (gsd-core/bin/lib/surface.cjs, introduced by ADR-0011 Phase 2) re-materializes a resolved Skill Surface profile to disk via applySurface. It currently hardcodes two artifact kinds (commands, agents) and re-derives their source directories via _findInstallSource / _findAgentsSource walk-up heuristics. The install and uninstall pipelines in bin/install.js each encode the same per-runtime artifact layout independently across ~14 install sites and ~6 uninstall sites. Bug #3659 surfaced the resulting drift: applySurface omits the skills kind for runtimes whose canonical layout is skills/gsd-<stem>/SKILL.md, so gsd-surface profile <name> leaves ~67 skill directories on disk under the install-time profile's footprint when the resolved profile should have pruned them — roughly 2.7k tokens per session on a measured workstation.
The root problem is the absence of a typed seam for "where does runtime R put artifact kind K." Three lifecycle sites (install, uninstall, surface) each independently encode this knowledge and drift independently.
Decision
- Add a Runtime Artifact Layout Module at
gsd-core/bin/lib/runtime-artifact-layout.cjsas the single owner of the per-runtime artifact-placement table. - The module requires
runtime-homes.cjsfor the canonical runtime enum and global config-dir resolution. It adds the artifact-kind axis on top. - Expose
resolveRuntimeArtifactLayout(runtime, configDir) → Layout. The returnedLayoutis a plain typed object —{ runtime, configDir, kinds: ArtifactKind[] }— with no I/O on resolution. - Each
ArtifactKindis{ kind: 'commands'|'agents'|'skills', destSubpath, prefix, stage }.stageis a function(resolvedProfile) → stagedDirthat closes over the per-runtime converter where one is needed (e.g.convertClaudeCommandToClaudeSkillfor theskillskind on Claude global). - The
kindsarray is empty for runtimes with no GSD surface (a hypothetical future runtime with no integration). Theskillskind is absent for runtimes that don't materialize skill directories (Cline; Gemini today). Thecommandskind is absent for runtimes that consume only the skills/agents layout (Claude global, Codex, etc.). - Per-runtime quirks live in the layout's record fields, not in caller branches:
- Hermes:
{ kind: 'skills', destSubpath: 'skills/gsd', prefix: 'gsd-' }— preserves the nested namespace from #2841. Note (#947): The original decision usedprefix: ''(bare stem) on the incorrect premise that theskills/gsd/category directory namespaced the leaf identifier in Hermes's loader. Research showed category dirs are purely organisational; dispatch is by the skillname:field. Thegsd-prefix was restored by #947 to match every other runtime. - Cline:
kinds: []— Cline resolves to zero kinds in Phase 1 (nocommandskind). - Gemini:
kinds: [ { kind: 'commands', destSubpath: 'commands/gsd', prefix: 'gsd-' } ]— no agents, no skills.
- Hermes:
applySurfacemigrates from(runtimeConfigDir, commandsDir, agentsDir, manifest, clusterMap)to(runtimeConfigDir, layout, manifest, clusterMap). Body collapses tofor (const kind of layout.kinds) _syncGsdDir(kind.stage(resolved), path.join(layout.configDir, kind.destSubpath), kind.kind)._findInstallSourceand_findAgentsSourceinsurface.cjsare removed. The layout owns source resolution.- Phase 2 (separate PR): install and uninstall paths in
bin/install.jsmigrate to iteratelayout.kinds. Per-runtime if/else branches for skill-directory creation/removal collapse to one layout-driven loop per pipeline. - Legacy-layout migrations (
bin/install.js:6710/:8402for the pre-nestedskills/gsd-*/flat layout;migrateLegacyDevPreferencesToSkillfor #2973) remain inside the Installer Migration Module (ADR-0008) and run before layout-driven copy. The layout module describes only the current canonical target — no historical kinds.
Initial Scope
Phase 1 should land the module and one consumer (the bug-#3659 fix):
- New
gsd-core/bin/lib/runtime-artifact-layout.cjs—resolveRuntimeArtifactLayout, the typedLayout/ArtifactKindshapes, and the runtime table covering every runtime currently enumerated inruntime-homes.cjs. surface.cjs:applySurfacemigrates to layout-driven iteration._findInstallSourceand_findAgentsSourcedeleted. Theskillskind is now iterated alongsidecommandsandagents— bug #3659 closed.commands/gsd/surface.mdandtests/surface-apply.test.cjsupdated to construct + passLayoutvalues.- New
tests/runtime-artifact-layout-*.test.cjscovering:- Per-runtime fixture table: each runtime maps to the expected
kinds[]shape. - Hermes
skills/gsdnested case. - Cline / Gemini "kind absent" cases.
- Source-root resolution (replacing the existing
surface.cjswalk-up tests).
- Per-runtime fixture table: each runtime maps to the expected
- Address the adjacent
readSurfacepartial-field silent-null bug noted in #3659 — out of scope here; track as a separateconfirmed-bugticket.
Phase 1 should not:
- Migrate the install/uninstall pipelines in
bin/install.jsin the same PR. That's a separate enhancement issue — same seam, larger blast radius. The layout module is dual-consumable from the start; convertingbin/install.jsis a sequenced follow-up. - Move the per-runtime skill converters (
convertClaudeCommandToClaudeSkill, etc.). They survive at their current file location as the stage adapters; the layout module references them. A future ADR may consolidate them into a Skill Conversion Module if a second consumer emerges.
Migration Inventory
New file
gsd-core/bin/lib/runtime-artifact-layout.cjs— module body + runtime layout table.
Files modified (Phase 1)
gsd-core/bin/lib/surface.cjs—applySurfacesignature change;_findInstallSource+_findAgentsSourceremoval.commands/gsd/surface.md— runbook updates the 3 sites that callapplySurfaceto first callresolveRuntimeArtifactLayout.tests/surface-apply.test.cjs— 5 call sites passlayoutinstead ofcommandsDir, agentsDir.
Files modified (Phase 2 — separate PR / separate issue)
bin/install.js— install path: 4 per-runtime skill-stage blocks collapse to one layout-driven loop.bin/install.js— uninstall path: 6 per-runtime skill-removal blocks collapse to one layout-driven loop.- Estimated reduction: ~250 lines.
Files not touched
runtime-homes.cjs— its narrow contract (runtime → global config dir / skills base) stays. The layout module is its sibling.install-profiles.cjs—stageSkillsForProfile,stageAgentsForProfileremain. The layout module'skinds[i].stageclosures call them.- The four skill converters at
bin/install.js:1622/1681/1792/2534— remain in place; layout closures reference them.
Interface sketch
// runtime-artifact-layout.cjs
/**
* @typedef {Object} ArtifactKind
* @property {'commands'|'agents'|'skills'} kind
* @property {string} destSubpath joined to layout.configDir
* @property {string} prefix 'gsd-' for all runtimes (incl. Hermes after #947)
* @property {(resolved) => string} stage returns staged dir path
*/
/**
* @typedef {Object} Layout
* @property {string} runtime canonical enum from runtime-homes.cjs
* @property {string} configDir caller-supplied (local or global scope)
* @property {ArtifactKind[]} kinds empty array = runtime has no gsd-* surface
*/
function resolveRuntimeArtifactLayout(runtime, configDir) { … }
Call-site shape:
// commands/gsd/surface.md (runbook), tests/surface-apply.test.cjs, future install/uninstall
const layout = resolveRuntimeArtifactLayout(runtime, runtimeConfigDir);
applySurface(runtimeConfigDir, layout, manifest, CLUSTERS);
applySurface body after migration:
function applySurface(runtimeConfigDir, layout, manifest, clusterMap) {
const resolved = resolveSurface(runtimeConfigDir, manifest, clusterMap);
for (const kind of layout.kinds) {
const staged = kind.stage(resolved);
const dest = path.join(layout.configDir, kind.destSubpath);
if (fs.existsSync(dest)) _syncGsdDir(staged, dest, kind.kind);
}
}
Consequences
- Bug #3659 becomes a fixture-table omission, not a forgotten if/else block. The layout-table test asserts every runtime's
kinds[]shape — forgetting theskillskind on Claude global would fail there. - The cross-runtime test matrix collapses from
N runtimes × 3 lifecycle verbs × 3 kinds(today: ~120 implicit assertion pairs) toN (layout table) + 3 (one per lifecycle verb iterating layout.kinds). - Adding a new runtime (recent example: Grok via commit
05316369) becomes one row in the layout table plus one fixture row in the test. The three lifecycle verbs pick it up automatically. - The four per-runtime skill converters become canonical adapters at the artifact-kind seam. Their existence is no longer accidental — they're the seam's content.
surface.cjsshrinks (_findInstallSource+_findAgentsSourceremoved). The walk-up heuristics — which were only ever-incidentally correct — are replaced by an explicit table.- Phase 2 (install/uninstall migration) shrinks
bin/install.jsby ~250 lines and removes a recurring class of bug: a new runtime added by a contributor who forgets to wire it through every install/uninstall branch. - Future architecture reviews should treat per-runtime artifact-placement knowledge added outside
runtime-artifact-layout.cjsas drift, parallel to how ADR-0011 made out-of-seam skill staging drift. - The Skill Surface Budget Module's leverage (typed sets of
{ skills, agents }) now extends all the way to disk through one seam rather than three independent re-materializations.
Open questions
- Whether the
skillskind'sstageclosure should accept the per-runtime converter as a parameter (preserving converter-as-pure-function purity) or import it directly. Implementation detail — settle in the PR. - Whether the layout module should expose a
listKinds(runtime)introspection helper for status/diagnostics surfaces, or keepLayoutas the only public type. Lean toward a single public type; add helpers only when a second consumer needs them. - Whether Phase 2 (install/uninstall migration in
bin/install.js) should land as a single follow-up PR or be split per pipeline. Lean toward single PR — the install and uninstall sides share the runtime branch structure and migrating only one introduces a temporary asymmetry inverse to today's. - The adjacent
readSurfacepartial-field silent-null fallback insurface.cjs:55-75(also flagged in #3659) is out of scope here — track separately. It's an independent shallow interface in the same module, not a runtime-layout concern.
References
- Confirmed bug:
#3659—applySurfacedoesn't prune~/.claude/skills/gsd-*/dirs - See
0011-skill-surface-budget-module.md— the Runtime Surface Module this seam serves - See
0008-installer-migration-module.md— legacy-layout migrations stay there - See
0005-sdk-architecture-seam-map.md— the seam map this module joins - Existing canonical sibling:
gsd-core/bin/lib/runtime-homes.cjs - Per-runtime skill converters this module references:
bin/install.js:1622(Copilot),:1681(Claude),:1792(Antigravity),:2534(Codex) - Hermes nested-skills layout rationale:
#2841
Implementation status
Phase 1 implementation landed on feat/3663-runtime-artifact-layout-module-phase-1-m:
gsd-core/bin/lib/runtime-artifact-layout.cjs— 15-runtime layout table (grok intentionally excluded),resolveRuntimeArtifactLayout(runtime, configDir, scope) → Layout, walk-upfindInstallSourceRoothelper.- Clarification: in this Phase 1 implementation, Cline resolves to zero kinds (
kinds: []), so it carries nocommandskind in the layout table. gsd-core/bin/lib/install-profiles.cjs— newstageSkillsForRuntimeAsSkills(srcCommandsDir, resolvedProfile, converter, prefix) → stagedDirhelper.gsd-core/bin/lib/surface.cjs—applySurface(runtimeConfigDir, layout, manifest, clusterMap)signature migration;_findInstallSource+_findAgentsSourcedeleted;_syncGsdDirextended to handle theskillskind via directory iteration.- Tests:
runtime-artifact-layout-resolve.test.cjs(16),runtime-artifact-layout-edge-cases.test.cjs(10),runtime-artifact-layout-stage.test.cjs(5),install-profiles-stage.test.cjs(+7 new),surface-apply.test.cjs(updated 5 call sites + new skills-kind test).
Phase 2 (separate issue #3664 — bin/install.js install/uninstall pipeline migration) is blocked on Phase 1 merge.
Amendment (2026-06-11): adding a runtime that rides the established layout is an addendum, not a new ADR
Maintainer governance decision. Registering an additional runtime that reuses the existing install machinery — the profile-marker-only install surface (ADR-58 / the config-adapter registry) plus a single skills kind in this module's layout table — is an enhancement governed by this ADR via this addendum, not a change that requires its own ADR. New design rationale (and a fuller amendment) is required only when a runtime introduces something this ADR has not already decided: a new install surface, a new ArtifactKind, or a layout quirk not expressible in the existing record fields (cf. the Hermes nested-namespace and Cline zero-kinds cases in the Decision section).
This codifies the lightweight path the project has used since Phase 1 for Codex, Copilot, Trae, Windsurf, Qwen, CodeBuddy, Cline, Kimi, et al., and makes the addendum-vs-new-ADR test explicit so contributors and reviewers stop re-litigating it per runtime.
Qualifying criteria (all three) for the addendum path
- No new install surface — the runtime maps to an existing
installSurfacevalue in the config-adapter registry (profile-marker-only, etc.);writesSharedSettings: false;finishPermissionWriter: nullor an existing writer. - No new artifact kind — the layout entry is composed only of the existing
commands/agents/skillskinds via the existingskillsKind(...)/ record-field machinery; no newArtifactKindshape. - Reuses the converter contract — per-runtime converters follow the established shape (see the agent-frontmatter contract below); only the path/name substitutions differ.
A runtime failing any of the three needs a fuller amendment here (or a new ADR) documenting the new surface/kind and its rationale.
Agent-frontmatter contract (normative)
Per-runtime agent converters emit a sanitized minimal frontmatter — name + description only — rebuilt through yamlIdentifier(name) / yamlQuote(toSingleLine(description)), matching convertClaudeAgentToTraeAgent / convertClaudeAgentToClineAgent / convertClaudeAgentToCodebuddyAgent. Claude-specific fields (tools:, color:, commented hook blocks) MUST NOT be passed through verbatim. A runtime that genuinely requires richer agent frontmatter must document the target schema in an amendment here and emit it deliberately — not arrive at pass-through by accident. (Recorded because PR #1021's first cut of convertClaudeAgentToQoderAgent fell through to full pass-through; it must be brought onto the sibling contract.)
Qoder (issue #860 / PR #1021) — first runtime recorded under this addendum
Qoder qualifies on all three criteria and is added as a layout-table entry, not a new design:
- Install surface:
profile-marker-only(config-adapter registry), identical shape to Trae/Windsurf — no statusline, nopackage.json, no sharedsettings.json/permission writes. - Layout:
case 'qoder'→[ skillsKind('skills', 'gsd-', 'convertClaudeCommandToQoderSkill', 'qoder', configDir) ];'qoder'added toALLOWED_RUNTIMES. - Home:
~/.qoder(orQODER_CONFIG_DIR), viaruntime-homes. - Converters:
convertClaudeToQoderMarkdown,convertClaudeCommandToQoderSkill,convertClaudeAgentToQoderAgent— the agent converter is subject to the contract above.
Path note: since ADR-457's .cts migration, the canonical module source is src/runtime-artifact-layout.cts (built to gsd-core/bin/lib/runtime-artifact-layout.cjs); the runtime enumeration is owned by src/runtime-homes.cts and src/runtime-config-adapter-registry.cts. The original body above predates that move; paths read accordingly.