Use superpowers to implement spec

[chapter37haptics]

Use superpowers to implement the spec from customized-domain branch

[chapter37haptics]

Codex Review (gpt-4.1-codex)

GATE: PASS — 4 findings, 0 critical

Findings (severity order)

1. Medium: Singular/plural DIR_PATTERN inconsistency
markdown.ts:361 accepts singular paths (/goal/, /decision/, /process/) via inferType, but link-extraction.ts:46 DIR_PATTERN only matches plural dirs (goals|decisions|processes). A link like [X](goal/slug) would get typed correctly but fail auto-linking.

Claude's take: Low real-world impact. Canonical directories are always plural. The existing codebase has the same pattern — inferType accepts /person/ but DIR_PATTERN only has people. Defensive fallback, not a bug.

2. Medium: doctor.ts graph_coverage excludes concept
The SQL clause adds goal/decision/process but still excludes concept, which is now a first-class developer entity.

Claude's take: Pre-existing gap, not a regression from this PR. concept was already excluded before these changes. Worth a follow-up.

3. Medium: writes_to drops VC entities (intentional)
brain-ops and signal-detector no longer declare people/, companies/, deals/, meetings/ in their writes_to frontmatter.

Claude's take: Intentional per spec — "VC types remain in the infrastructure but the agent's behavioral layer stops triggering on them."

4. Low: Test coverage gaps
Missing: singular-dir extraction tests, wikilink tests for decisions/ and processes/, regression tests confirming old dirs still match after DIR_PATTERN expansion.

Verification Evidence

• bun run typecheck: PASS (exit 0)
• 485 tests across 7 affected test files: 0 failures
• bun run verify (CI pre-test gate): PASS

Recommendation

Ship as-is. Finding #1 is a pre-existing pattern in the codebase, #2 is a pre-existing gap, #3 is intentional. Finding #4 (additional test coverage) is polish, not a blocker.

---

🤖 Review by Codex CLI (codex exec, reasoning effort: high, 53k tokens) + Claude synthesis

[chapter37haptics]

Claude's Evaluation of Codex Review Findings

Verified each finding against the actual codebase before deciding what to do.

Finding	Verdict	Reasoning
#1 Singular/plural DIR_PATTERN	No action	Pre-existing pattern. inferType accepts /person/ but DIR_PATTERN only has people. The singular forms are defensive fallbacks for directories that canonically use plural names.
#2 doctor.ts excludes concept	No action (initially)	Pre-existing gap, not a regression. concept was already excluded before this PR.
#3 writes_to drops VC entities	No action	Intentional per spec: "VC types remain in the infrastructure but the agent's behavioral layer stops triggering on them."
#4 Test coverage gaps	No action	We have 1 wikilink test for goals/ proving DIR_PATTERN works for new entries. Additional singular-dir and wikilink tests are polish, not blockers.

[chapter37haptics]

Codex Second Opinion on Claude's Dismissals

Asked Codex to evaluate whether Claude's dismissals of the original 4 findings hold up.

Finding	Codex Grade	Codex's Take
#1 Singular/plural DIR_PATTERN	PARTIAL	Claude is right on low impact, but the existing person/people mismatch is "precedent for technical debt, not proof this is safe."
#2 doctor.ts excludes concept	PARTIAL	"Pre-existing" is factually plausible but weak — this PR did modify that exact query and expanded entity typing. Leaving out concept keeps health metrics misaligned with the new first-class entity framing. Worth fixing.
#3 writes_to drops VC entities	VALID	Intentional and internally consistent. Only caveat: it is a behavior change and should be called out as such.
#4 Test coverage gaps	PARTIAL	"One wikilink test proves DIR_PATTERN works" is too broad a claim. Real risk is low-to-moderate, not zero.

Action taken: Fixed finding #2 — added concept to the doctor.ts graph_coverage SQL clause (commit 5856aa2). Codex was right that touching the query and leaving concept out was a missed opportunity with a one-line fix.

Other findings: dismissals stand. #1 and #4 are real but low-risk technical debt, #3 is intentional.

[chapter37haptics]

Codex Spec-vs-Tests Audit

Asked Codex to grade each spec requirement against the actual test suite.

Spec Requirement	Grade	Notes
types.ts: add goal/decision/process to PageType + ALL_PAGE_TYPES	TESTED (loose)	No assertion on ALL_PAGE_TYPES count (25)
markdown.ts inferType(): map goals/, decisions/, processes/	TESTED (tight)	3 direct path tests + longest-prefix test
inferType longest-prefix: projects/x/decisions/y -> decision	TESTED (tight)	Regression would fail
link-extraction.ts DIR_PATTERN: add goals/decisions/processes	TESTED (tight)	Both markdown-link + wikilink syntax covered
VC types stay in infrastructure	TESTED (loose)	people + companies checked, not full legacy set
DIR_PATTERN used by ENTITY_REF_RE and WIKILINK_RE	TESTED (tight)	Both syntaxes directly tested

Untested but not actionable (skill-layer agent behavior, not unit-testable):
MECE boundaries, debug trail placement, page template schema, execution order

Action taken: Added ALL_PAGE_TYPES.length === 25 assertion to pin the exact type count (commit incoming).