Commit Graph

6 Commits

Author SHA1 Message Date
Tom Boucher
7f13ee5373 enhance(#2793): ADR-2782 — reviewer lane becomes a declared capability surface (#2809)
* docs(#2793): add ADR-2782 — reviewer lane capability surface

Design lock for epic #2782. Declares a reviewer lane as capability data
rather than a core patch across three unrelated surfaces.

Key decisions:
- D2 transport discriminator (spawn | openai-http) — a survey of all
  twelve lanes found three that are HTTP endpoints with no binary, which
  invalidated the single-invoke-shape draft.
- D4 the reviewer body is optional and absent-safe at every layer.
- D5 a fourth executable-surface disclosure class covering the lane
  binary or host AND its egress payload classes.
- D6 handler is a closed first-party enum, upholding ADR-1016; the
  consequence — third-party lanes are data-only — is stated plainly.
- D7 probe kinds wider than existence, and every probe bounded.

Amends ADR-857, ADR-894, ADR-1016, ADR-1244. Also records the
D7/D8-extended-by-ADR-1244 marker on ADR-857 that ADR-1244 D8 promised
but never added.

Closes #2793

* docs(#2793): address orthogonal review findings on ADR-2782

Two blockers from the isolated adversarial pass:

- D5 disclosed the spawn binary but not its args, reopening the #1459
  bug class already fixed for MCP servers (binary python3 + args
  -c <program>). args are now disclosed and signature-bound.
- hostConfigKey resolves from .planning/config.json, which is mutable
  after consent with no integrity check, so a lane consented against
  localhost could be silently redirected to a remote host by an
  ordinary PR. The resolved host is now consent-bound and re-verified
  on the invocation path; a mismatch blocks the lane.

Majors and spec gaps:

- D4 gains an explicit-selection carve-out. Absent-safe governs
  discovery, never a lane the user named; the current selector records
  that as info, which Phase 1 now corrects.
- D4 gains a warning delivery channel.
- D6 enumerates the handler closed-enum members; a closed enum whose
  membership is left to the implementing phase is not closed.
- D6 records aider and plandex as concrete lanes the vocabulary cannot
  express, rather than claiming sufficiency it did not verify.
- D2 gains evidenceClass, requiresBinaries, promptBudgetKey for
  per-lane divergence that was only prose, and motivates the
  one-member outputChannel enum.
- reviewer.requires renamed requiresBinaries — it collided with the
  envelope requires (capability deps) at a different nesting depth.
- Antigravity two-level timeout: Context cited it then dropped it;
  now explicitly delegated to the handler.
- D9 gains a per-key ownership table, including three keys that stay
  central because they are policy across lanes, not lane properties.
- Phase table maps every decision D1-D9 to a delivering phase; D6
  handler modules and D5 invocation-time re-verification were
  previously unclaimed.
- American English per house style.
2026-07-28 22:26:26 -04:00
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
Tom Boucher
bcc5a6d1ba fix(#1634): honor capability hook matcher and node-prefix command (#1638)
* fix(#1634): honor capability hook matcher and node-prefix command

Capability hook install (applyCapabilitySharedEdits) wrote each settings.json
hook entry with no `matcher`, so a tool-scoped hook fired on every tool (a
fail-closed guard could then block the whole session), and emitted a bare
single-quoted script path so a .js-family hook from a git/tarball source
without +x failed with Permission denied on every matching call.

- Pass through an optional declared `matcher` (entry-level sibling of `hooks`);
  absent => omitted (match-all), so existing shipped capabilities are unchanged.
- Validate `matcher` in the declaration (non-empty string, no control chars).
- Emit `node <quoted-path>` for .js/.cjs/.mjs hooks (mirrors first-party);
  .sh and others keep the bare quoted path (unchanged).

Root cause: the manifest hook schema (validator rule C4) was {event, script}
only with no matcher, and applyCapabilitySharedEdits never read or wrote one;
the command used shellSingleQuote(absScript) with no node prefix.

Regression tests fail-first on both defects (matcher dropped; bare path) and
pass after the fix; #1460 command assertions updated for the node prefix.

* chore(#1634): backfill changeset pr:1638

* fix(#1634): resolve lint and windows CI failures

- validator: replace the control-character range regex with a char-code loop.
  The literal /[\x00-\x1f\x7f]/ tripped ESLint's no-control-regex rule; char
  codes are equally precise and lint-clean. Behavior unchanged (still rejects
  matchers containing ASCII control characters incl. DEL).
- test: gate the executable-bit precondition on POSIX. Windows fs does not
  honor POSIX write modes (a 0o644 write reads back as 0o666), so the
  precondition is meaningless there and failed the windows-latest lane. The
  node-prefix assertion — the actual fix — is platform-independent and still
  runs everywhere.

* docs(#1634): amend ADR-894 for optional lifecycle hook matcher

The `role: "feature"` `hooks[]` entry now carries an optional `matcher`
(settings.json tool-scoping pattern: exact/pipe/wildcard/regex). Document
the field in the §2 schema table and record a Grilling-amendments entry:
the install path projects a declared matcher onto the emitted settings.json
hook entry (absent = match-all, so shipped capabilities are unchanged), and
per-runtime matcher projection (ADR-857 D8) stays a separate concern. This
amendment ships with the fix that introduced the field rather than as a
follow-up.

* docs(#1634): record WINDOWS-POSIX-MODE-BIT-ASSERT defect in CONTEXT.md

Capture the CI failure pattern from #1634/PR #1638 so it is not repeated: a
test that writes a file with a POSIX mode and then asserts statSync().mode
& 0o777 === <octal> passes on macOS/Linux but fails on windows-latest
(Windows fs does not honor POSIX write modes — reads back 0o666). Added as a
machine-greppable DEFECT predicate (symptom/examples/detect/fix-forward/
prevention) next to DEFECT.WINDOWS-TEST-PORTABILITY, with the fix-forward:
gate the mode-bit precondition on process.platform !== 'win32' and keep the
platform-independent behavioral assertion running everywhere.
2026-06-23 21:49:43 -04:00
Tom Boucher
eb051ea696 feat(#1123,#1124): enforce duplicate-producer invariant + fail-loud loadCentralConfigKeys in gen-capability-registry (#1131)
Closes #1123
Closes #1124
Refs #857
2026-06-12 16:51:02 -04:00
Tom Boucher
5f48a42514 docs(#1022): resolve step-vs-gate model question — steps additive, gates block, mode self-gates (#1025)
Record the resolution of #1022 (surfaced scoping the §5.6 ui-phase cutover):
a step is purely additive and never halts the host; host-blocking preconditions
are gates (blocking/onError:halt) — no hook-model change. Runtime/mode context
(auto/chain vs manual) self-gates in the skill, not via when (config-only).

§5.6 decomposes into the existing plan:pre step (ui-phase, self-gates on
frontend + pipeline) + a new plan:pre gate (frontend-and-no-UI-SPEC → halt,
when: workflow.ui_safety_gate) that blocks planning in manual mode — preserving
the "run /gsd:ui-phase first" UX (maintainer call: pipelines-only auto-fire).
The render-hooks dispatch template grows to handle gates, not just steps.

Recorded in ADR-894 (clarification) + CONTEXT.md
(RULESET.CAPABILITY.step-additive-gate-blocks). Unblocks the §5.6 cutover.

Closes #1022

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-10 22:51:31 -04:00
Tom Boucher
197d6bffe2 docs(#894): ADR-894 Capability declaration format + registry generation (#895)
* docs(#894): ADR-894 Capability declaration format + registry generation

ADR-857 rollout phase 3a (design-only). Resolve ADR-857's deferred open
question — the on-disk Capability declaration format — as a reviewable design
ADR before any generator code.

Specifies: the capabilities/<id>/capability.json folder layout (migration-staged
ownership — declarations reference existing stems until the phase-6 move); the
capability.json schema for role:feature (skills/agents/hooks/federated config/
loopHooks) and role:runtime (the six closed projection-primitive axes); the 12
named Loop Extension Points; the gen-capability-registry.cjs generator design
(validation + cross-capability invariants + --write/--check drift gate, mirroring
gen-inventory-manifest); the generated capability-registry.cjs shape (by-id /
by-skill / by-loop-point indexes + requires-closure); and a full worked example
(the UI capability: ui-phase + ui-review + agents + config + two loop hooks).

No code — design artifact only; the generator build, federated config loader
(3b), and loop seam (3c) implement against this contract.

Closes #894

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

* docs(#894): amend ADR-894 with grilled capability declaration format

Stress-tested the declaration format before merge; the format changed
materially. Amendments:
- loopHooks[] -> three typed arrays (steps/contributions/gates), each with its
  own shape (step: ref+produces/consumes; contribution: fragment+into agent-role;
  gate: check+blocking).
- Add the Loop Host Contract (§3): each step publishes its points, agent roles,
  and core artifacts so the generator validates hooks against reality, not
  trusted strings.
- requires = capability ids only (host implicit); add tier-monotone invariant;
  drop the requires:["plan"] error from the example.
- Config federation = atomic move: a migrated key leaves the central schema in
  the same PR; presence in both is a collision (invariant stays).
- One registry, role-partitioned indexes (feature indexes vs runtimes index).
- Rework the UI worked example to the split-array shape (2 steps + 1 gate) +
  a contribution illustration.

Adds a "Grilling amendments" section recording the six changes.

* docs(#894): amend ADR-894 with round-2 grilling (operational reality)

Second design-grill round, folded in before merge:
- Loop Host Contract is GENERATED from structured workflow markers
  (<loop-point>/<agent-role>/<loop-artifact>) via gen-loop-host-contract.cjs —
  it can't drift from the real workflows.
- Hook activation `when`: cheap deterministic config-level gating evaluated by
  loop.render-hooks; deeper phase-context applicability self-gates inside the
  dispatched skill (no phase-context vocabulary to keep honest).
- `tier` is the source of install-profile + cluster membership; profiles and
  clusters are generated from tier + requires-closure (/gsd:surface operates on
  capabilities) — collapses ADR-857's multiple toggle systems.
- Gate `check` = query | declarative-predicate | agentVerdict; agentVerdict is
  forced advisory; only deterministic checks may block.
- byLoopPoint ordering is materialized in the registry; render-hooks filters the
  active set + renders. Same-capability hooks degrade gracefully when an entry
  step self-gates.
- Rollout: registry-only until atomic per-feature cutover (no double-execution
  with still-inlined workflow features).

Updates the Grilling amendments / Consequences / Alternatives / Open questions
sections; reworks the UI example with `when`.

---------

Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-08 18:51:20 -04:00