Repository navigation
Fix context rot and a risky skill step found by a prompt audit - #3
Merged
Merged
Conversation
A prompt audit of the agent-facing surface found four lines that no longer match the repo: - skills/AGENTS.md said the davila7 mirror was synced at v0.2.0 with v0.3.0 pending. davila7/claude-code-templates#911 merged v0.3.0 on 2026-09-17 and the vendored copy matches its hash. Drop the hand-kept sync state; the sentence before it already says how to tell that a re-sync is pending. - app/AGENTS.md's boundary map left out RawSourceView and ErrorView, and the two source-of-truth modules (site-routes.ts, site-definition.ts) the other docs point agents at. - AGENTS.md justified the Lighthouse wording against a version the reader never saw. Drop that sentence; the rule it explains stays. - content/AGENTS.md said "There is no slogan", a leftover of the removed slogan. Say what to do instead: do not add one. Claude-Session: https://claude.ai/code/session_01GEVZ7EuuDgvN3ULSo6qiVF
doc-references only checked that the names app/AGENTS.md cites exist, so the map could leave a component out and stay green (it did: RawSourceView and ErrorView). Assert that every top-level component and every composable is named, with a check that the assertion can fail. Run against the map on main, it flags both missing components. The root AGENTS.md test map says so. Claude-Session: https://claude.ai/code/session_01GEVZ7EuuDgvN3ULSo6qiVF
… skill as 0.3.1 The context-rot step told the agent to verify that every referenced command "still runs". Read literally, that runs deploy, publish, or migration scripts during an audit the skill declares read-only. Check instead that the script or binary is defined, and run only the side-effect-free ones (lint, typecheck, unit tests). Bump the version to 0.3.1 in the SKILL.md frontmatter and both marketplace.json entries, and register the new hash in skill-version.test.ts. The davila7 mirror needs a re-sync once this ships. Claude-Session: https://claude.ai/code/session_01GEVZ7EuuDgvN3ULSo6qiVF
4 of 5 tasks
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.
Summary
A prompt audit of the agent-facing surface (SKILL.md, the AGENTS.md set, REVIEW.md, the Claude Code config) against the current Claude generation. The surface is clean on the patterns that date prompts: no capitalized pressure language, no "think step by step", no pinned models, no tools or request code. What it found is context rot, two lines phrased against earlier versions, and one skill instruction that is unsafe when read literally.
skills/AGENTS.mdsaid thedavila7/claude-code-templatesmirror was at v0.2.0 with v0.3.0 pending; #911 merged v0.3.0 on 2026-09-17 and the vendored copy matches its hash.app/AGENTS.md's boundary map left outRawSourceView,ErrorView,site-routes.ts, andsite-definition.ts.AGENTS.mdjustified the Lighthouse wording against a version the reader never saw;content/AGENTS.mdsaid "There is no slogan", a leftover of the removed slogan, now stated as "do not add one".doc-references.test.tsnow asserts thatapp/AGENTS.mdnames every top-level component and every composable, plus a check that the assertion can fail. Run against the map onmain, it flagsErrorViewandRawSourceView, the rot fixed here.marketplace.jsonentries (edited by hand, under the principle-09 deny rules), with the new hash registered inskill-version.test.ts.Test plan
pnpm lint,pnpm format:check,pnpm typecheckpnpm generatethenpnpm test: 14 files, 85 tests passclaude plugin validate --strict .mainversion ofapp/AGENTS.md(mutation check)Follow-up after merge (outside the tree)
davila7/claude-code-templatesmirror to the 0.3.1SKILL.md.SKILL.mdis now 20,217 of the 20,480 bytesskill-spec.test.tsallows. The next addition needs a trim or a move to a reference file.Generated with Claude Code
https://claude.ai/code/session_01GEVZ7EuuDgvN3ULSo6qiVF