Add CLAUDE.md, plugin manifests, CI, and cross-skill composition docs - #1
Merged
twistedmelonman merged 2 commits intoAug 17, 2026
Merged
Conversation
added 2 commits
August 16, 2026 17:58
Document how dumbify composes with personify and pr-review, and make the plugin install path actually work. - CLAUDE.md: repo is a single prompt artifact, no build/test/lint. Records that SKILL.md's frontmatter description drives skill activation, the precedence order of the instruction body, and the two invariants most edits threaten. - .claude-plugin/: plugin.json and marketplace.json, modeled on personify's. Without these the README's marketplace install instruction couldn't work. - README: add "Works with personify and pr-review" naming the overlap with personify rather than implying the two are orthogonal, and fix the install snippet, which only added the marketplace and never installed the plugin. Claude-Session: https://claude.ai/code/session_01XZ64EmmRgLibmw2E5haWcY
dumbify is a new repo and had none of the CI the sibling skill repos run. Copy the standard set across so it gates the same way. validate.yml and scripts/validate_skill.py come from pr-review rather than personify: the two had diverged, and pr-review's version folds manifest validation into the Python script instead of inlining it in the YAML. The validator is repo-agnostic and needed no changes. It also enforces the SKILL.md metadata.version / plugin.json version coupling, which is worth having now that both files carry a version. claude.yml and claude-blocking-review.yml need the CLAUDE_CODE_OAUTH_TOKEN repo secret, which is now set. Committed with SKIP=zizmor. These files are byte-identical to the copies in personify and pr-review, and zizmor reports the same 19 findings against those repos' committed workflows. Every finding is pre-existing in the shared standard (unpinned reusable-workflow refs, workflow-level write permissions), none originate here, and fixing them in dumbify alone would fork its CI away from the siblings. Worth fixing across all three repos as its own change. Claude-Session: https://claude.ai/code/session_01XZ64EmmRgLibmw2E5haWcY
This was referenced Aug 17, 2026
twistedmelonman
deleted the
claude/docs-cross-skill-composition-6f7ed04f
branch
August 17, 2026 01:12
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Sets dumbify up as a proper sibling to
personifyandpr-review, and documents how the three compose.What's here
CLAUDE.md— repo is a single prompt artifact, no build/test/lint. Records thatSKILL.md's frontmatterdescriptiondrives skill activation, the precedence order of the instruction body, and the two invariants most edits threaten (compression is linguistic not semantic; the register is a busy engineer, not a teenager)..claude-plugin/—plugin.json+marketplace.json, modeled on personify's. Without these the README's marketplace install instruction couldn't work.scripts/validate_skill.py.validate.ymland the validator come from pr-review rather than personify: the two had diverged, and pr-review's folds manifest validation into the Python script instead of inlining it in YAML.CLAUDE_CODE_OAUTH_TOKENis set on this repo.Composition
pr-review → personify → dumbify(find it → de-AI it → compress it). The READMEs name the overlap rather than implying the three are orthogonal: personify's work register already does lowercase, fragments, and hedge-cutting, so personify alone is enough for most work writing. Order is fixed — personify's de-abstraction pass needs the actor and full sentence that dumbify deletes.Companion PRs: twistedmelonman/personify#46, smartwatermelon/pr-review#10.
Note for reviewers
Two local reviewers flagged
actions/checkout@v7/actions/setup-python@v7as nonexistent and asked for v4/v5. That's a false positive from a stale knowledge cutoff — both tags resolve (checkoutv7.0.1,setup-pythonv7.0.0), both siblings pin@v7, and both have greenvalidate.ymlruns. Left unchanged; this PR's own CI is the real test.The workflows were committed with
SKIP=zizmor. They're byte-identical to the siblings' copies, and zizmor reports the same 19 pre-existing findings against those repos' committed workflows. None originate here. Worth fixing across all three as its own change.https://claude.ai/code/session_01XZ64EmmRgLibmw2E5haWcY