Skip to content

docs: reconcile skill catalog with reality, file ADR-0008#11

Draft
cooper (czxtm) wants to merge 2 commits into
mainfrom
claude/admiring-mayer-703904
Draft

docs: reconcile skill catalog with reality, file ADR-0008#11
cooper (czxtm) wants to merge 2 commits into
mainfrom
claude/admiring-mayer-703904

Conversation

@czxtm

Copy link
Copy Markdown
Member

Summary

While importing skills/ADRs from this repo into darkmatter/.github's org instructions, found that docs/catalog.md has drifted from the actual skills/ directory:

  • 15 catalog rows have no matching skills/<name>/ directory (coding-standards, hl-funding-analysis were deliberately removed in 5347d8d without a catalog update; the rest — browser-use, caveman, caveman-review, compress, dispatching-parallel-agents, executing-plans, frontend-design, neon-postgres, subagent-driven-development, systematic-debugging, tdd-workflow, verification-before-completion, writing-plans — have no git history at all, i.e. were never built). Several real skills (brainstorming, using-superpowers, ui-component-architecture, vercel-react-best-practices) reference these phantom names as prerequisites/peers, so anyone following the catalog hits a dead end.
  • The real skills/tdd/ directory was undocumented; the catalog instead documented a fictional tdd-workflow "legacy redirect" that never existed as a directory.
  • skills/README.md was linked twice from the root README.md as "the skill format spec" but never existed.
  • ADR-0008 (decouple telemetry concerns) was fully written up in the darkmatter/obsidian wiki (as its own 0006-decouple-telemetry-concerns.md, accepted 2026-05-22) but never filed in this repo's docs/adr/ — the source of truth the wiki explicitly points back to. Backported it here.

Changes

  • docs/catalog.md: move the 15 phantom entries to a new "Known gaps" section instead of silently deleting the record that they're referenced elsewhere; add the real tdd skill; drop the fictional tdd-workflow row; fix dangling precedence-note references.
  • skills/README.md: new — skill format spec (frontmatter, disable-model-invocation, scripts//reference/ conventions, validation/cataloging steps).
  • docs/adr/0008-decouple-telemetry-concerns.md: new — backport of the wiki's already-accepted decision.

Test plan

  • bash scripts/validate-skill.sh passes across all skills (unaffected — no skill directories were added/removed, only skills/README.md, which the validator correctly skips since it only walks directories)
  • Verified via git log that coding-standards/hl-funding-analysis were deliberately removed and the other 13 "known gap" names have no directory history at all

Generated by Claude Code

docs/catalog.md advertised 15 skills with no matching skills/<name>/
directory (coding-standards and hl-funding-analysis were deliberately
removed in 5347d8d without a catalog update; the rest were never built).
Several real skills (brainstorming, using-superpowers, ui-component-
architecture) reference these phantom names as prerequisites, so anyone
following the catalog would hit a dead end.

- Move phantom entries to a "Known gaps" section instead of silently
  deleting the record that they're referenced elsewhere
- Add the real `tdd` skill (was undocumented) and drop the fictional
  `tdd-workflow` entry, which never existed as a directory
- Add skills/README.md (skill format spec), which the root README
  linked to twice but never existed
This decision has been in effect since 2026-05-22 and was already fully
written up as an ADR in the darkmatter/obsidian wiki, but was never filed
in this repo's docs/adr/ — the source of truth the wiki explicitly points
back to. Org-level instructions (.github CLAUDE.md/AGENTS.md) referenced
it only as an unnumbered "OTel-only observability" bullet. Backport the
wiki's content here as ADR-0008 so it has one canonical source.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants