Files
msd-core/docs/adr/3524-cjs-sdk-hard-seam.md
Tom Boucher 67a9243cf1 chore(#2356): make the ADR index a generated artifact and enforce ADR lifecycle invariants (#2367)
* chore: rebuild ADR index as a generated artifact and enforce lifecycle invariants

The ADR index in docs/adr/README.md was hand-maintained with nothing checking
it, and had drifted to 40 of 65 ADRs. The absent rows included the entire
capability family (857/894/959/1016/1143/1213/1244) and ADR-1239 (EoS) itself,
so the decisions a reader most needed were the ones they could not find.

Make the index a derived artifact, matching the repo's existing generated-file
idiom (lint:generated-sync), and enforce the corpus' lifecycle invariants:

- scripts/gen-adr-index.cjs generates the index between markers and validates
  the status vocabulary (Accepted/Proposed/Superseded/Legacy/Retired),
  successor links, id/filename agreement, and supersession symmetry.
- Wire --check into lint:generated-sync so drift fails CI.

Correct the lifecycle metadata the gate surfaced, without flipping any status:

- ADR-1239 (EoS) declared it subsumed ADR-1016/58/3660/894; none recorded it.
  Add reciprocal "Subsumed by" pointers + dated amendments. Subsumption keeps
  the target Accepted -- these are live adapters, not dead decisions.
- ADR-857/894 carry dated status caveats: they read Proposed while the
  capability system shipped and epic #857 is closed. Ratification is a
  maintainer act and is deliberately left open.
- Link ADR-0005/0007/0012/3524 -> ADR-0174 and ADR-0010 -> ADR-0009; record
  the reciprocal Supersedes on ADR-0009.
- ADR-218 declared itself "ADR-0175" -- an unfinished rename.
- The 0011 PRD moves from the non-canonical "Draft" to "Legacy".

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* test: capture stderr via spawnSync; record ADR-0010 draft supersession

Two fixes surfaced by the first gsd-test run and by regenerating the index:

- tests/adr-index-gate.test.cjs used execFileSync, which only surfaces stderr
  through the thrown error on non-zero exit. The `--write` path exits 0 while
  reporting outstanding violations on stderr, so the helper always saw ''.
  spawnSync captures both streams on both outcomes.
- The hand-maintained index recorded 0010-skill-surface-budget-module.md as
  "earlier draft superseded by ADR-0011" while the file itself still said
  Proposed. Deriving the index from the files would have dropped that
  assertion and resurrected a superseded draft as a live decision, so it is
  recorded at its source, with the reciprocal Supersedes on ADR-0011.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* fix: drop the dead sdk/ model-catalog candidate retired by ADR-0174

src/model-catalog.cts resolved model-catalog.json through three candidates, the
second being sdk/shared/model-catalog.json three levels up. That was the legacy
source-repo fallback kept by the #3288 fix ("check the co-located path FIRST,
before the legacy source-repo path").

ADR-0174 then retired the @opengsd/gsd-sdk package boundary and deleted the sdk/
tree (11918dcc3), so the candidate can no longer resolve in any layout: a source
repo has no sdk/, and an install layout points it at ~/.claude/sdk/shared/, which
the installer never writes -- the original #3288 bug. It was dead weight implying
a package boundary this repo no longer has.

No test depends on it: the #3288 regression tests in tests/install.test.cjs write
their own synthetic old-path fixture and assert it throws.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* docs: ratify nine shipped ADRs; record why ten others stay Proposed

The corpus carried 19 Proposed ADRs, most describing architecture that had
already shipped. A Proposed label on live architecture tells contributors and
agents the decision is an unbuilt idea -- the capability system and EoS were
both being misread that way.

Audited all 19 against the shipped tree and GitHub. Each candidate flip then had
to survive two independent reviewers instructed to refute it.

Ratified Proposed -> Accepted, each with a dated Ratification section carrying
the verified evidence (file:line, symbols, tests, issue state):

  857  capability system      894  declaration format   1244 capability ecosystem
  1577 injection boundary     1610 size-budget ratchet  1990 existing-code onboarding
  15   cross-AI convergence   22   plan-drift guard     0011 default reviewers

Held ten, each now carrying a "Why this is still Proposed" section naming the
blocker and its unblock condition, so the audit is not repeated:

  2264 its own headline acceptance criterion is unmet in the tree
  230  live branch protection contradicts the decided spec (1 approval, not 2)
  660  the namesake release/<version> re-cut is manual, not automated
  959  issue #2346 is approved and plans its graduation as its own ADR
  1213 the shipped writer's return shape differs from the decided interface
  443  the orchestrator override path has no live caller
  1143 / 1606 each states its own bar for acceptance; neither is met
  612 / 1671 legitimately open

Shipped code proved necessary but not sufficient: eight ADRs had every named
module, symbol, and test present with their epics closed, and still failed the
bar. That lesson is written into README.md's ratification procedure.

Also corrected ADR-857's "Supersedes (generalizes)" to "Subsumes": taken
literally it would have marked two live seams dead -- ADR-0011 (surface.cts:348)
and ADR-58 (runtime-artifact-install-plan.cts:82). Both keep Accepted status and
gain Subsumed-by pointers.

Index: Active 39->48, Proposed 19->10, Superseded/Legacy 7. 65 total.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* fix: harden gen-adr-index against hostile titles and non-ADR filenames (#2356)

Three findings from the pre-PR orthogonal security review, all confirmed:

- An ADR title containing the literal ADR-INDEX:END marker was emitted verbatim
  into its table cell, relocating the splice boundary so the NEXT --write
  spliced against the wrong marker and truncated README.md. Titles now render
  through cellText(), which escapes pipes and angle brackets -- making an HTML
  comment (and any other HTML) unformable from ADR-authored text.
- A docs/adr/*.md without a numeric prefix crashed on match(...)[1] of null.
  Such a file is also invisible to the index -- the very failure this gate
  exists to prevent -- so it is now reported as a naming-convention violation
  naming the file and the fix.
- Tests leaked their mkdtemp dirs. They now use helpers.createTempDir/cleanup
  via t.after(); helpers.cleanup carries the Windows-EBUSY retry budget that a
  raw fs.rmSync lacks (caught by local/no-raw-rmsync-in-tests).

Adds five regression tests: marker hijack, HTML injection, pipe cell-break,
non-conforming filename, and splice stability across repeated writes.

Refs #2356

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* fix: close two gate false-passes; read ## Supersedes sections (#2356)

Second round of confirmed findings from the pre-PR orthogonal code review. Both
false-passes matter more than a false-fail: a gate that silently misses a
violation is worse than no gate, because it is trusted.

- A relation field mixing a link with a bare id silently dropped the bare claim:
  the check tested `rel.links.length` (does this field have ANY link?) instead
  of whether THAT id was linked. `Supersedes: [ADR-0001](...), ADR-0011` passed
  clean -- accepting exactly the ambiguous bare reference the rule forbids. Now
  each bare id is checked against the ids actually linked in the same field, so
  a repeat in trailing prose stays quiet while an unlinked claim is flagged.
- The ratification guard (`statusToken !== 'Accepted'`) skipped BOTH relation
  directions, which killed the IN check entirely: `supersedes.in` is only ever
  populated on an ADR whose status IS `Superseded`, so a dangling `Superseded by
  X` where X never claims it always passed. The guard now applies to OUT only --
  a prospective claim must not obligate its target, but an ADR's statement about
  ITSELF is always owed a reciprocal.
- Fixing that surfaced a parser gap: ADR-0174 declares its supersessions in a
  `## Supersedes` table SECTION, not a header field, and headerBlock() stops at
  the first `##`. The repo's best-documented supersession was invisible. Section
  form is now parsed for both relations.
- Replaced a vacuous test: the em-dash negation case passed whether or not
  NEGATED_RELATION_RE matched (a mutation to /$^/ survived). It now carries a
  link that would create a failing asymmetric relation if negation did not fire.

Also removes docs/adr/9401-test-target.md -- a synthetic fixture a reviewer
created in the worktree while reproducing a finding, swept in by `git add -A`.

Adds regression tests for each: mixed link+bare, linked-and-repeated-in-prose,
dangling superseded-by from a non-Accepted ADR, and the ADR-0174 section shape.

Refs #2356

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* fix: escape backslashes before pipes in the ADR index cell renderer (#2356)

CodeQL js/incomplete-sanitization (high) on scripts/gen-adr-index.cjs: cellText()
escaped `|` -> `\|` without first escaping the backslash. Markdown's escape
character is the backslash, so the input `\|` became `\\|`, which renders as a
literal backslash followed by an UNESCAPED pipe -- re-opening the cell break the
pipe escape exists to prevent. Order is load-bearing: escape the escape
character first, then everything that emits one.

Same class as the index-marker hijack fixed earlier: ADR-authored text breaking
out of the cell it is rendered into.

Adds a regression test asserting a `\|`-bearing title leaves exactly the row's
own 5 unescaped delimiters and cannot forge a Status cell. Uses split(/\r?\n/)
per local/no-crlf-fragile-split -- a literal "\n" split is CRLF-fragile on the
Windows CI leg.

Refs #2356

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-17 10:51:58 -04:00

24 KiB
Raw Blame History

CJS↔SDK hard seam — one source of truth per Shared Module

  • Status: Superseded by ADR-0174 (2026-05-23); originally Proposed (2026-05-14)
  • Date: 2026-05-14
  • Tracking issue: #3524
  • Related PRD: docs/prd/3524-cjs-sdk-hard-seam.md
  • Extends: ADR-0005 (seam map) — adds the Shared-Module Source Policy to the seam family
  • Defers to: ADR-0001 (Dispatch Policy Module), ADR-0003 (Model Catalog Module), ADR-0004 (Planning Workspace Module), ADR-0006 (Planning Path Projection Module), ADR-0009 (Shell Command Projection Module — post-Phase 3–4, also subsuming superseded ADR-0010)

We decided to harden the boundary between the CJS tooling layer (gsd-core/bin/lib/*.cjs) and the SDK (sdk/src/**/*.ts) by making every Module that is conceptually shared between the two runtimes have exactly one hand-authored source of truth and at most one generated artifact per runtime. The trigger is the recurring drift bug class — #1535, #1542, #2047/#2052, #2638/#2655, #2653/#2670, #2687/#2706, #2798/#2816, #3055/#3116, #3523 — each of which was a fix landing on one side without the other.

The precedent shape is already in the repo. sdk/scripts/gen-command-aliases.ts emits sdk/src/query/command-aliases.generated.ts and gsd-core/bin/lib/command-aliases.generated.cjs from one TypeScript source. sdk/scripts/check-command-aliases-fresh.mjs is the CI freshness gate. The two consuming sides are pure Adapters over the generated artifact. This ADR generalizes that pattern to the other Shared Modules and forbids the hand-synced-pair anti-pattern that produced #3523.

Decision

1. Shared-Module Source Policy

A Shared Module is any Module whose Interface is consumed identically by both the CJS toolset and the SDK. The CONTEXT.md domain glossary already calls these out — e.g. STATE.md Document Module is explicitly typed as "Shared CJS/SDK pure transform Module."

For every Shared Module:

  1. Exactly one hand-authored source of truth. Lives at sdk/src/<module-name>/ as TypeScript when the Module has behavior, or sdk/shared/<module-name>.manifest.json when the Module is pure data.
  2. Generated artifacts only. The CJS-side file is gsd-core/bin/lib/<module-name>.generated.cjs and is emitted mechanically. It is never hand-edited.
  3. Per-Module freshness check. A CI script sdk/scripts/check-<module>-fresh.mjs re-runs the generator and fails if the emitted artifact differs from the committed one. Precedent: check-command-aliases-fresh.mjs.
  4. Per-Module drift lint (when the source is data, not a generator output). Precedent: scripts/lint-shell-command-projection-drift.cjs. The lint asserts the canonical-owner invariants that aren't captured by file-equality.
  5. Hand-synced pairs are forbidden. A pre-merge lint-shared-module-handsync.cjs greps gsd-core/bin/lib/ for non-.generated.* files whose basename matches a sdk/src/query/<same-name>.ts source and fails the build unless the pair is explicitly allow-listed.

2. Module-indexed canonical-owner table

The table below indexes by Module, not by physical layer. Each row names the source of truth, the emitted artifacts, the Adapter sites, and either the new ADR section that defines the Module or the existing ADR that already owns it.

Module Status Source of truth Generated artifacts Adapters
STATE.md Document Module New under this ADR (Phase 1) — see CONTEXT.md "STATE.md Document Module" sdk/src/state/index.ts (promoted from sdk/src/query/state-document.ts) sdk/src/query/state-document.generated.ts, gsd-core/bin/lib/state-document.generated.cjs bin/lib/state.cjs and sdk/src/query/state*.ts import the generated form
Configuration Module New under this ADR (Phase 2) — definition added to CONTEXT.md as part of Phase 2 sdk/src/config/index.ts plus data manifests sdk/shared/config-schema.manifest.json and sdk/shared/config-defaults.manifest.json sdk/src/query/config-schema.generated.ts, gsd-core/bin/lib/config-schema.generated.cjs, gsd-core/bin/lib/configuration.generated.cjs bin/lib/config.cjs, bin/lib/core.cjs:loadConfig, sdk/src/config.ts
Workstream Inventory Module (Builder) Amended under this ADR (Phase 3) — Builder split documented in CONTEXT.md update sdk/src/workstream/builder.ts (pure projection from directory entries + STATE.md text + plan scan results → typed inventory) sdk/src/query/workstream-inventory-builder.generated.ts, gsd-core/bin/lib/workstream-inventory-builder.generated.cjs Per-side fs Readers (workstream-inventory.cjs sync, workstream-inventory.ts async) call the Builder. Readers stay hand-authored because the fs idiom legitimately differs.
Project-Root Resolution Module New under this ADR (Phase 4) — short CONTEXT.md entry, behavior already de-facto shared sdk/src/project-root/index.ts gsd-core/bin/lib/project-root.generated.cjs bin/lib/core.cjs (findProjectRoot, findEffectiveRoot), sdk/src/helpers.ts
Frontmatter Module Conditional (Phase 3, only if drift catalogue confirms pair duplication) sdk/src/frontmatter/index.ts gsd-core/bin/lib/frontmatter.generated.cjs Existing handler call sites
Plan Scan Module Conditional (Phase 3 or later) sdk/src/plan-scan/index.ts gsd-core/bin/lib/plan-scan.generated.cjs Phase/roadmap routers
CJS Command Router Adapter Module Amended under this ADR (Phase 5). Existing Module (per CONTEXT.md) is extended so the per-family handlers map delegates to the SDK runtime bridge in-process instead of to parallel CJS handler implementations. sdk/src/query-runtime-bridge.ts (already exists) + per-family delegate emitter gsd-core/bin/lib/cjs-command-router-adapter.cjs (existing, ~40 lines) plus per-family handlers maps that require('../../sdk/dist/query-runtime-bridge.cjs') and call QueryRuntimeBridge.execute() bin/gsd-tools.cjs and the seven bin/lib/*-command-router.cjs files are the consumers. Per-family CJS handler files (state.cjs, verify.cjs, init.cjs, etc.) shrink to delegates or are deleted once the SDK handler is the only implementation.
Command-Alias Module Already sealed by this pattern's precedent — sdk/scripts/gen-command-aliases.ts + check-command-aliases-fresh.mjs No change No change No change
Dispatch Policy Module Defer — see ADR-0001 (and its 2026-05-05 SDK Runtime Bridge amendment) n/a n/a n/a
Model Catalog Module Defer — see ADR-0003; the sdk/shared/model-catalog.json manifest already follows the source-of-truth policy n/a n/a n/a
Planning Workspace Module Defer — see ADR-0004; withPlanningLock, workstream pointer policy, lock semantics stay where they are n/a n/a n/a
Planning Path Projection Module Defer — see ADR-0006; SDK is canonical, CJS path resolution converges via Phase 4 if any divergence is found n/a n/a n/a
Shell Command Projection Module (incl. platform fs + subprocess after Phase 3–4 expansion) Defer — see ADR-0009; this Module is the canonical owner for platformWriteSync, platformReadSync, platformEnsureDir, execGit, execNpm, execTool, probeTty, normalizeContent n/a n/a n/a
Skill Surface Budget Module Defer — see ADR-0011 (accepted, not the 0011-superseded draft) n/a n/a n/a

3. Out-of-seam Modules (per-runtime, no shared source)

These remain CJS-only. Drift cannot occur because there is no SDK counterpart. If any later needs an SDK port, that port is a new enhancement, not a parallel implementation.

  • bin/lib/graphify.cjs
  • bin/lib/gsd2-import.cjs
  • bin/lib/schema-detect.cjs
  • bin/lib/fallow-runner.cjs
  • bin/lib/intel.cjs
  • bin/lib/drift.cjs
  • bin/lib/installer-migrations.cjs (installer runtime is CJS-native; SDK consumes via sdk-package-compatibility.ts Adapter)

4. Per-side I/O Adapters legitimately differ

The per-side state Adapter, verify Adapter, and similar handlers are not in the Shared-Module table. CJS callers use synchronous fs/exec; SDK callers use async I/O and the SDK observability decorators. The pure transforms behind them (parsing, projection, normalization) are extracted into Shared Modules per the table above; the I/O remains per-side. Golden parity tests in sdk/src/golden/ pin observable behavior across the seam.

5. Enforcement (per existing repo precedents, not new conventions)

Drift is blocked at three layers, each modeled on an existing in-repo script:

  1. Per-Module freshness check — sdk/scripts/check-<module>-fresh.mjs, one per Shared Module in the table. Precedent: check-command-aliases-fresh.mjs.
  2. Per-Module drift lint (when invariants are not pure file-equality) — scripts/lint-<module>-drift.cjs, one per data-manifest-backed Module. Precedent: lint-shell-command-projection-drift.cjs.
  3. Hand-sync pair lint — scripts/lint-shared-module-handsync.cjs rejects any pair of files at gsd-core/bin/lib/<name>.cjs and sdk/src/query/<name>.ts (or sdk/src/<name>.ts) that are neither generated artifacts nor on an explicit allow-list. This blocks the #3523 anti-pattern at PR time.

CODEOWNERS extends to sdk/src/<module>/ for each Shared Module. Architecture-team review is required for changes to a source of truth.

A top-of-file banner is auto-inserted by each generator into the emitted .generated.cjs / .generated.ts files. Banner pattern follows the existing command-aliases.generated.* files: a header noting "GENERATED FILE — Source: …". No additional banner tooling is introduced.

6. New CONTEXT.md entries added by this ADR's phases

  • Configuration Module (added during Phase 2): Module owning config load, legacy-key normalization, defaults merge, and explicit on-disk migration for .planning/config.json. Interface: loadConfig(cwd) → MergedConfig (pure read, no disk write); normalizeLegacyKeys(parsed) → { parsed, normalizations[] } (idempotent, returns the list of normalizations applied for migration logging); mergeDefaults(parsed) → MergedConfig; migrateOnDisk(cwd) → MigrationReport (explicit, opt-in, called only by the installer and by gsd-tools migrate-config). Invariants: never mutates disk inside loadConfig; legacy top-level keys (branching_strategy, sub_repos, multiRepo, depth) are normalized into their canonical nested locations in the returned value; defaults come from the shared config-defaults.manifest.json.
  • Project-Root Resolution Module (added during Phase 4): Module owning project-root and effective-root resolution heuristics including own-.planning detection, parent-sub_repos traversal, legacy multiRepo, and .git-ancestor fallback.
  • Workstream Inventory Module — Builder split (CONTEXT.md amendment during Phase 3): the existing Module entry gains a sub-paragraph noting that the pure projection logic is the source of truth and the per-side Reader Adapters are hand-authored over the generated Builder.
  • CJS Command Router Adapter Module — runtime-bridge delegation (CONTEXT.md amendment during Phase 5): the existing Module entry gains a paragraph noting that the per-family handlers map delegates to QueryRuntimeBridge.execute() in-process via require('../../sdk/dist/query-runtime-bridge.cjs'). Per-side CJS handler files (state.cjs, verify.cjs, etc.) that previously held parallel implementations are reduced to delegates or deleted once their SDK counterpart is the only remaining implementation. CJS-only Module handlers (graphify, gsd2-import, schema-detect, fallow-runner, intel, drift) keep their in-process CJS implementations because no SDK counterpart exists.

Consequences

  • The hand-synced-pair anti-pattern that produced #3523 becomes impossible to merge. The lint-shared-module-handsync.cjs gate rejects any new pair that is not generated. The check-<module>-fresh.mjs gates reject any edit to a generated file that is out of sync with its source.
  • The seam vocabulary stays inside the existing CONTEXT.md / LANGUAGE.md frame. No new layer labels ("shared core", "shared data"); the unit of seam ownership is the Module, as it already is everywhere else in this repo.
  • No new build tooling is introduced. The generator pattern is the existing gen-command-aliases.ts shape. No dual CJS+ESM bundler, no package.json exports subpath change, no tsup/rollup decision.
  • Each phase ships one Shared Module. The smallest phase (STATE.md Document Module) ships first because both files are already character-identical — the deletion test passes on contact. The trigger bug class (#3523) is closed in Phase 2 by the Configuration Module. The seam becomes a real wall in Phase 5 when the CJS routers stop holding parallel handler implementations.
  • CJS dispatch collapses onto the SDK runtime bridge. Once Phase 5 lands, every canonical command running via gsd-tools executes the same SDK handler that gsd-sdk query executes — in-process, not subprocess. The per-side state/verify/init/phase/roadmap/validate handler implementations in CJS are replaced by thin delegates over QueryRuntimeBridge.execute(). The result-shape contract is preserved ({ exitCode, stdoutChunks, stderrLines } per the Query CLI Output Module, ADR-0001).
  • Existing ADRs are deferred to, not restated. Planning Path Projection (ADR-0006), Model Catalog (ADR-0003), Planning Workspace (ADR-0004), Dispatch Policy (ADR-0001), Shell Command Projection (ADR-0009) remain authoritative for their domains. The new ADR adds Shared-Module Source Policy, the per-Module entries above, and the CJS Command Router Adapter Module amendment.
  • Per-side I/O Adapter divergence is preserved at the runtime-bridge boundary. The CJS router's sync execution model is preserved: QueryRuntimeBridge.execute() exposes a sync entry point for CJS callers (or, when the underlying SDK handler is async, the bridge runs an in-process event loop step). No subprocess hop is added. Async SDK call sites continue to use the async bridge directly.
  • Enforcement reuses existing scripts. Three new lint/check primitives, all modeled on scripts already in scripts/ and sdk/scripts/. CI wiring follows the existing precedent.

Out of scope

  • Migrating CJS-only Modules (graphify, gsd2-import, schema-detect, fallow-runner, intel, drift) to SDK handlers — each is its own enhancement.
  • Sync→async migration of CJS state/verify Adapters — leaves the per-side Adapter shape intact, which is the point.
  • Defining a Verify Module before the verify surface has a shared Interface — that is precondition work for a future enhancement, not this one.

Amendments

(Append-only. Use a dated header when the decision evolves.)

2026-05-23 — validate.ts → verify.cjs generator pattern (issue #6)

Three pure helpers from sdk/src/query/validate.ts Check 8 are now generated into gsd-core/bin/lib/validate.generated.cjs via sdk/scripts/gen-validate.mjs, following the same I/O adapter pattern established by PR #154 (issue #4):

Generator: sdk/scripts/gen-validate.mjs Artifact: gsd-core/bin/lib/validate.generated.cjs Freshness check: sdk/scripts/check-validate-fresh.mjs CI: .github/workflows/test.yml — "SDK generated validate artifact drift check"

Three drift items resolved (issue #6):

  1. W007 activeDiskPhases — verify.cjs Check 8 previously iterated diskPhases (which includes archived milestone phases via forEachArchivedPhaseToken) for the W007 check. Archived phases absent from the current ROADMAP produced false W007 warnings. Fix: W007 now iterates activeDiskPhases (from collectDiskPhases() only, without forEachArchivedPhaseToken), matching validate.ts Check 8 behavior.

  2. phaseVariants() normalization — verify.cjs Check 8 used parseInt(p).padStart(2,'0') for disk-existence and roadmap-membership checks, which drops letter suffixes (e.g. "3B" → "03" instead of "03B"). Phase dirs with letter-suffix padding mismatches (ROADMAP "3B", disk "03B-foo") produced false W006 and W007. Fix: both checks now use phaseVariants(p) from the generated module, which returns the full normalized Set including letter-suffix forms.

  3. W006 unchecked-phase variant skip — verify.cjs Check 8 built notStartedPhases with raw + parseInt-padded forms (drops letter suffix). phaseVariants() is now used instead, so unchecked entries like "3B" correctly suppress W006 for "03B" (and vice versa).

phaseVariants extraction note: phaseVariants is defined as a closure inside validateHealth in the compiled output (not a module-level export). It is extracted via brace-balanced source-text parsing from sdk/dist/query/validate.js, the same technique used for escapeRegex extraction in gen-phase-lifecycle-policy.mjs. The function is deterministic and pure: no closures over external state, no side effects.

Parity tests: tests/health-validation.test.cjs (folds former 6-validate-cjs-drift-regression, consolidation epic #1969) — 5 tests (all GREEN after fix, all RED on pre-fix origin/main). Covers each drift item with concrete fixtures:

  • Drift 1: two milestone archives (v1.0 old, v1.1 active); v1.0 phase absent from ROADMAP.
  • Drift 2: ROADMAP "01A", disk "1A-foo" — padding mismatch.
  • Drift 3: ROADMAP "3B", disk "03B-foo" — zero-padded letter-suffix mismatch.

Allowlist: scripts/shared-module-handsync-allowlist.json — verify.cjs entry updated to reference the generator and freshness check. Classification remains cooperating-sibling (verify.cjs is still a full implementation; only Check 8 helpers are generated).

Extension — issue #26: W005/W006-archived/I001 generator migration

PR #3479 fixed three false-positive classes in sdk/src/query/validate.ts. PR #3806 hand-ported the three fixes to gsd-core/bin/lib/verify.cjs but did not route them through the generator — meaning they could drift again. Issue #26 closes this gap by extending gen-validate.mjs (introduced in this amendment above) to also extract and export the W005/W006-archived/I001 items.

Four additional exports added to validate.generated.cjs (issue #26):

  1. phaseDirNameRe (W005) — The PHASE_DIR_NAME_RE constant /^\d{2,}(?:\.\d+)*-[\w-]+$/ is now a named export from validate.ts and extracted by gen-validate.mjs. verify.cjs Check 6 consumes phaseDirNameRe from the generated artifact instead of an inline copy. Reproducer: mkdir -p .planning/phases/999.1-foo → zero W005 (previously fired with the \d{2} two-digits-only regex before PR #3806 / PR #3479).

  2. PHASE_TOKEN_FROM_DIR_RE (W006-archived) — The regex constant previously inline in verify.cjs's forEachArchivedPhaseToken() and collectDiskPhases(). Extracted from the module-level const in the compiled output. verify.cjs inline copy removed.

  3. MILESTONE_ARCHIVE_DIR_RE (W006-archived) — The regex constant previously inline in verify.cjs's listMilestoneArchiveDirs(). Extracted the same way. verify.cjs inline copy removed. Together PHASE_TOKEN_FROM_DIR_RE and MILESTONE_ARCHIVE_DIR_RE ensure the archive-walking logic uses the same patterns as validate.ts.

  4. canonicalPlanStem (I001) — The top-level helper function previously inline in verify.cjs Check 7. Extracted via extractTopLevelFunction() (brace-balanced parser). verify.cjs inline copy removed. Fix: 68-01-scaffolding-PLAN.md correctly matches 68-01-SUMMARY.md — both reduce to 68-01 via canonicalPlanStem().

W006-archived coverage note: Issue #26 describes W006-archived as "RELATED TO but DISTINCT FROM" PR #156's W006 fix. Investigation confirmed both fixes are ALREADY in verify.cjs (from PR #3806). The gap was generator coverage: the regex constants used by forEachArchivedPhaseToken were inline copies with no generator protection. This amendment closes that gap by extracting them. No new behavioral fix is required — the generator pattern extension is the deliverable.

validate.ts change: PHASE_DIR_NAME_RE promoted from inline anonymous regex to a named export const so it appears as an extractable identifier in the compiled ESM output.

Extraction methods used:

  • extractConstRegExp() (new in gen-validate.mjs) — handles const and export const single-line RegExp assignments. Used for phaseDirNameRe, PHASE_TOKEN_FROM_DIR_RE, MILESTONE_ARCHIVE_DIR_RE.
  • extractTopLevelFunction() (new in gen-validate.mjs) — brace-balanced parser for top-level named function declarations. Used for canonicalPlanStem.

Parity tests: tests/health-validation.test.cjs (folds former 26-w005-w006-i001-cjs-drift-regression, consolidation epic #1969) — 7 tests.

  • W005: no false positive for 999.1-foo; W005 still fires for single-digit prefix.
  • W006-archived: no false W006 for phase archived under milestones/v1.0-phases/; unit tests for MILESTONE_ARCHIVE_DIR_RE and PHASE_TOKEN_FROM_DIR_RE export and behavior.
  • I001: no false I001 when long-stem PLAN matches short-stem SUMMARY via canonicalPlanStem; I001 still fires when there is genuinely no SUMMARY; unit test for canonicalPlanStem export.

Cross-references: issue #26 cures the same false-positive scenarios as issue #6 but for the W005/W006-archived/I001 check paths. The artifact validate.generated.cjs now covers all six drift surfaces originally identified across both issues. This completes the validate.ts ↔ verify.cjs migration scope for generator-pattern coverage.

2026-05-23: Phase * cooperating-sibling retirement — I/O adapter pattern for phase lifecycle (issue #4)

Context: Issue #4 revealed that phase.cjs:cmdPhaseComplete was non-idempotent: every call blindly incremented Completed Phases by 1 and computed Progress without a 100% clamp. The SDK's phase-lifecycle.ts already contained the correct idempotent implementation ("Root cause 1 fix" block ~line 1644), but CJS had no way to share it because the mutation handlers are async I/O-bound and Section 4's "I/O stays per-side" rule prevents direct sharing.

Decision: Apply the I/O adapter pattern (Section 4) to the pure-computation kernel inside phase-lifecycle.ts:

  1. Three new generator scripts extract pure helpers from the phase family:

    • sdk/scripts/gen-phase.mjs → gsd-core/bin/lib/phase.generated.cjs (pure helpers: isCanonicalPlanFile, describeNonCanonicalPlans)
    • sdk/scripts/gen-phase-lifecycle.mjs → gsd-core/bin/lib/phase-lifecycle.generated.cjs (pure helpers: deriveProgressFromRoadmap, clampPercent)
    • sdk/scripts/gen-phase-lifecycle-policy.mjs → gsd-core/bin/lib/phase-lifecycle-policy.generated.cjs (14 pure policy helpers: generatePhaseSlug, computePhaseDirectory, buildPhaseRoadmapEntry, etc.)
  2. phase.cjs:cmdPhaseComplete is migrated to use deriveProgressFromRoadmap + clampPercent from the generated artifact. It reads the freshly-updated ROADMAP synchronously, derives the completed-phase count from Complete-row matching (idempotent), and passes it through clampPercent to prevent >100% Progress.

  3. Freshness checks: check-phase-fresh.mjs, check-phase-lifecycle-fresh.mjs, check-phase-lifecycle-policy-fresh.mjs added. CI adds three corresponding drift-check steps alongside the existing generated-artifact drift checks.

  4. Allowlist: phase.cjs ↔ phase.ts remains cooperating-sibling because phase.cjs still owns the CJS sync mutation handlers (phaseAdd, phaseInsert, phaseRemove, phaseComplete) that are legitimately I/O-bound and cannot be extracted via .toString(). The allowlist justification is updated to reflect the new generator consumption.

What this is NOT: The full async mutation handlers (phaseAdd, phaseInsert, phaseRemove, phaseComplete) are inherently async I/O-bound and are NOT generated — per Section 4. This amendment only extracts the pure-computation kernel.

Open drift bugs remaining: Issue #6 (phasePlanIndex drift) and issue #26 (phase.add inline ROADMAP update drift) are separate bug reports; they are referenced here for traceability but not fixed in this amendment. Future amendments should note when those are resolved.