Relocate this repo's ADRs and invocation.md from docs/ to .agents/,
and update the CLAUDE.md reference.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- Revert Sketch back to Prototype to mesh with the /prototype skill and
existing vocabulary, keeping the generalized function-first definition.
- Move the `## Notes` block from the top of the map to the end.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Decouple the skill from engineering so it plans any domain (e.g. course
content) without specializing toward it. Domain specifics live in the map,
not the skill body.
- Add an optional `## Notes` block to the map: declares the domain + skills
to consult, plus freeform standing preferences. Loaded into every session,
honored by every session, accrues as it emerges.
- Rename the Prototype ticket type to Sketch and define it by function —
raise the fidelity of the discussion with a cheap concrete artifact to
react to (/prototype is one medium among outline/rough-take/stub).
- Note that validation is the thread through all three types, not a type.
- Resolve step now consults skills named in Notes.
The session machinery (one ticket per session, claiming, parallelism,
handoff) is unchanged — it follows from the AI agent's smart/dumb zones, not
the domain.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add a header note clarifying that link-skills.sh is a dev-only script for
maintainers, not a supported installer, and that modifications or requests
for modification will not be approved.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Each smell now reads what-it-is → how-to-fix, telegraphic style — the fix
is the actionable half a reviewer needs, kept to a clause.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Folds Refactoring ch.3 "Bad Smells in Code" into the in-progress `review`
skill. The skill stays two-axis: smells layer into the Standards axis as a
fixed, always-on baseline alongside a repo's documented standards, rather
than a third axis.
Because the baseline is needed on every run it lives inline in SKILL.md.
Each smell is one line — its name carries the definition (the names are
strong leading words) plus a diff-specific cue. Curated to ~12 high-signal
smells; vague ones (Comments, Loops) and tooling-caught ones (Long
Parameter List) are deliberately dropped.
Two rules bind the baseline: a documented repo standard always overrides
it, and every smell is reported as a judgement call, never a hard
violation. The Standards sub-agent prompt now carries the baseline and
these rules.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Refactoring belongs to the review stage, not the implementation loop.
Remove the refactor rule and the now-orphaned refactoring.md (its home is
the review skill), and point the loop's rule at the review skill so it
isn't re-added.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The red → green → refactor loop is anchored by leading words the model
already holds, so the step-by-step Workflow mostly restated the loop and
duplicated the horizontal-slicing anti-pattern. Drop the Workflow and
per-cycle checklist; fold their durable idea (vertical slices / tracer
bullets) into Anti-patterns and a short Rules-of-the-loop list.
Introduce **seam** as the leading word for where tests go, collapsing the
old Philosophy "public interfaces" prose and the Planning "confirm
interface / behaviors" handshake into one rule: test only at pre-agreed
seams, confirmed with the user before any test is written.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Tests whose assertion is recomputed the way the code computes it pass by
construction and give zero confidence. Add it as a peer of the existing
implementation-coupling anti-pattern: a Philosophy principle, a per-cycle
checklist gate, and a BAD/GOOD example pair in tests.md. Includes a patch
changeset.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Replace numeric ticket ids (#1, #2…) with short, token-efficient
dash-case slugs (e.g. relational-db) that read as mini-titles. Update
the Structure example, the next-ticket selection rule (lowest-numbered →
first in document order), the optional-arg wording, and the Handoff
copy-paste examples accordingly.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add `publish: npx changeset tag` to the release workflow and enable
tagging for private packages in the changesets config, so merging a
Version PR creates a git tag and GitHub release.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add changesets for tracking and versioning skill changes:
- package.json (private) + @changesets/cli and @changesets/changelog-github
- .changeset/config.json using the GitHub release-style changelog
- release workflow that opens a version PR on push to main (no npm publish)
- initial changeset recording recent skill updates
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add a Pruning directive to run the no-op test sentence by sentence and
delete whole failing sentences rather than trim words. References the
existing no-op test as single source of truth — no restated definition.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- Validate the fixed point resolves and diff is non-empty before spawning sub-agents
- Single-source the tooling-enforced and two-axis rationale (point to Why two axes)
- Fix rerank contradiction: summarize worst issue per axis, not across axes
- Drop dead "Read X, then read the diff" preambles from both briefs
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
