bchanot/claude
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

refactor(audits): route all report writes to .claude/audits/

Browse Source 
Before: SEO.md, GEO.md, HARDEN.md, VALIDATE.md, BUGS-FOUND.md landed
at project root. After: all five go to .claude/audits/. Covers both
dispatcher write paths and the dispatcher bash commands that parse
each report (test -s, grep score, wc) — otherwise the dispatcher
would look for the file at the old location.

- skills/seo,geo,harden,validate,code-clean — write paths + console
  summaries ("Report: .claude/audits/X.md")
- skills/harden,validate — bash parsing commands (test/grep/wc) aligned
- agents/seo-analyzer,validator-analyzer,code-cleaner — agent-side refs
- agents/validator-analyzer frontmatter description updated

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
bastien 2026-04-23 16:06:54 +02:00
parent 32ccceb548
commit 7b57b2e091
8 changed files with 41 additions and 44 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
agents/code-cleaner.md
View File

@ -141,7 +141,7 @@ Do NOT call the `/refactor` skill — invoke the agent directly.
If cleanup reveals actual bugs (not style issues — real defects):
- Append each bug to `BUGS-FOUND.md` at project root:
- Append each bug to `.claude/audits/BUGS-FOUND.md` (run `mkdir -p .claude/audits` first):
```
## [date] Bug found during code-clean
- **File**: <file:line>
@ -178,7 +178,7 @@ REFACTORED:
SKIPPED (user decision):
- <item> — <reason>
BUGS FOUND: <N> (logged to BUGS-FOUND.md)
BUGS FOUND: <N> (logged to .claude/audits/BUGS-FOUND.md)
TESTS: passing / no test suite / <failures>
```
@ -192,7 +192,7 @@ TESTS: passing / no test suite / <failures>
- No "while we're here" scope creep. Only fix approved items.
- Exported/public API symbols require explicit per-item user confirmation
before deletion — even if they appear unused.
- Bugs go to BUGS-FOUND.md, not fixed in this workflow.
- Bugs go to .claude/audits/BUGS-FOUND.md, not fixed in this workflow.
- If the codebase has no tests and the changes are non-trivial,
warn the user about the risk before executing.
- No plugin check (lightweight skill, not an orchestrator).

2
agents/seo-analyzer.md
View File

@ -877,7 +877,7 @@ SEO AGENT RESULT (depth: <LOCAL|FULL>)
```
**If standalone `/seo` on a project without `/geo`**: write/update
`SEO.md` at project root. Structure matches classic format, with §7
`.claude/audits/SEO.md` (run `mkdir -p .claude/audits` first). Structure matches classic format, with §7
(GEO) marked as "Not audited — run /geo for GEO/AI optimization".
```markdown

11
agents/validator-analyzer.md
View File

@ -1,6 +1,6 @@
---
name: validator-analyzer
description: Web standards audit agent — W3C HTML validity (validator.nu), W3C CSS validity (jigsaw.w3.org), WCAG 2.1 accessibility (axe-core, pa11y, WAVE). Dispatched from /validate. Produces scored VALIDATE.md report with concrete diffs for auto-fixable issues and user actions for judgment-required fixes. Complementary to /harden (security), /seo (indexability), /geo (AI extraction).
description: Web standards audit agent — W3C HTML validity (validator.nu), W3C CSS validity (jigsaw.w3.org), WCAG 2.1 accessibility (axe-core, pa11y, WAVE). Dispatched from /validate. Produces scored .claude/audits/VALIDATE.md report with concrete diffs for auto-fixable issues and user actions for judgment-required fixes. Complementary to /harden (security), /seo (indexability), /geo (AI extraction).
tools: Read, Edit, Write, Bash, Grep, Glob, WebFetch
---
@ -287,7 +287,7 @@ RGAA 4.1 (French public sector).
---
## STEP 4 — Score + VALIDATE.md
## STEP 4 — Score + .claude/audits/VALIDATE.md
### Scoring
@ -306,7 +306,7 @@ Base 100. Deductions :
Clamp to [0, 100].
### Report structure — write to `<PROJECT_ROOT>/VALIDATE.md`
### Report structure — write to `<PROJECT_ROOT>/.claude/audits/VALIDATE.md` (run `mkdir -p .claude/audits` first)
```markdown
# Validation Report — <project name>
@ -455,7 +455,7 @@ READY TO APPLY — awaiting dispatcher confirmation
- **Single agent, narrow scope.** W3C HTML + W3C CSS + WCAG 2.1 only.
Drop anything else (meta tags, JSON-LD, perf, security, generic linting).
- **Degrade gracefully.** Missing tools → fall back to static checks.
Never fail hard. Always produce VALIDATE.md, even in degraded mode.
Never fail hard. Always produce .claude/audits/VALIDATE.md, even in degraded mode.
- **Framework awareness.** For SPA/JS frameworks (Next/Astro/Vite/
SvelteKit/Nuxt), validate built output (`dist/`, `_site/`, `build/`,
`out/`), not JSX/TSX source. Note "Validated against built HTML at
@ -464,8 +464,7 @@ READY TO APPLY — awaiting dispatcher confirmation
STOP, return control via `READY TO APPLY`.
- **Cite evidence.** Every finding : `file:line` + tool output quote.
Empty findings or hand-waving = bug.
- **One report.** `VALIDATE.md` at project root (or `docs/VALIDATE.md`
if convention exists). On re-run, move previous content to a
- **One report.** `.claude/audits/VALIDATE.md`. On re-run, move previous content to a
`## Historique` section — do not overwrite silently.
- **External validators are authoritative.** If validator.nu disagrees
with `html-validate`, trust validator.nu. If jigsaw disagrees with

2
skills/code-clean/SKILL.md
View File

@ -7,7 +7,7 @@ description: |
Trigger: "code-clean", "clean up the code", "remove dead code",
"enforce code style", "cleanup", "nettoyage du code", "code hygiene".
For targeted refactoring without audit → use /refactor instead.
For bug fixes discovered during cleanup → logged to BUGS-FOUND.md, not fixed here.
For bug fixes discovered during cleanup → logged to .claude/audits/BUGS-FOUND.md, not fixed here.
argument-hint: <file, directory, or blank for entire project>
disable-model-invocation: false
allowed-tools:

4
skills/geo/SKILL.md
View File

@ -37,9 +37,9 @@ $ARGUMENTS
## Note on integration
If `SEO.md` already exists at project root, the geo-analyzer will
If `.claude/audits/SEO.md` already exists, the geo-analyzer will
merge its findings into that file's `§7 — Optimisation GEO / IA`
section (rather than writing a separate `GEO.md`). This keeps a
single consolidated report when both /seo and /geo have been run.
If no `SEO.md` exists, the agent writes `GEO.md` at project root.
If no `.claude/audits/SEO.md` exists, the agent writes `.claude/audits/GEO.md` (run `mkdir -p .claude/audits` first).

27
skills/harden/SKILL.md
View File

@ -8,7 +8,7 @@ description: |
(.htaccess, nginx.conf, netlify.toml, vercel.json, _headers, _redirects,
wrangler.toml). Dispatches the seo-analyzer agent with a STRICT scope
filter — no meta/OG/JSON-LD/sitemap/CWV/headings/alt/i18n noise.
Produces HARDEN.md at project root.
Produces .claude/audits/HARDEN.md.
Trigger: "harden", "web hardening", "ssl audit", "https audit",
"hsts", "csp", "security headers", "http to https", "redirect audit",
"htaccess audit", "404 page", "canonical audit", "transport security",
@ -366,7 +366,7 @@ Agent(
"READY TO APPLY — awaiting dispatcher confirmation" at the end.
Do NOT apply any Edit/Write — the dispatcher handles STEP 3.
OUTPUT — write to <PROJECT_ROOT>/HARDEN.md :
OUTPUT — write to <PROJECT_ROOT>/.claude/audits/HARDEN.md (run `mkdir -p .claude/audits` first) :
# Web Hardening Report — <project_name>
@ -504,18 +504,18 @@ cached (now-READY) SSL Labs result.
## STEP 2 — Verify output
```bash
test -s HARDEN.md && wc -l HARDEN.md || echo "MISSING HARDEN.md"
test -s .claude/audits/HARDEN.md && wc -l .claude/audits/HARDEN.md || echo "MISSING .claude/audits/HARDEN.md"
```
If missing or empty :
```
⚠️ seo-analyzer did not produce HARDEN.md. Options:
⚠️ seo-analyzer did not produce .claude/audits/HARDEN.md. Options:
A) Retry with same scope
B) Downgrade to LOCAL and retry (if FULL failed on network)
C) Abort
```
Extract the score and critical-alert count from HARDEN.md for the console summary.
Extract the score and critical-alert count from `.claude/audits/HARDEN.md` for the console summary.
---
@ -523,9 +523,9 @@ Extract the score and critical-alert count from HARDEN.md for the console summar
Skip this step if MODE=audit.
If MODE=fix and HARDEN.md ends with `READY TO APPLY — awaiting dispatcher confirmation` :
If MODE=fix and `.claude/audits/HARDEN.md` ends with `READY TO APPLY — awaiting dispatcher confirmation` :
1. Parse the `## 8. Fix bundle` section from HARDEN.md.
1. Parse the `## 8. Fix bundle` section from `.claude/audits/HARDEN.md`.
2. Group by file. For each group, show the combined diff to the user.
3. Ask :
```
@ -539,15 +539,15 @@ If MODE=fix and HARDEN.md ends with `READY TO APPLY — awaiting dispatcher conf
A) Apply all
B) Review each diff before applying
C) Apply only Critique severity
D) Abort — keep HARDEN.md as audit report
D) Abort — keep .claude/audits/HARDEN.md as audit report
```
4. On `A` : apply each bundle via Edit (targeted old_string/new_string,
never full-file Write on shared templates).
5. On `B` : for each diff, show and ask yes/no/skip.
6. On `C` : filter to Critique-only, then behave as `A`.
7. On `D` : stop, leave HARDEN.md untouched.
7. On `D` : stop, leave `.claude/audits/HARDEN.md` untouched.
After applying : append a `## 10. Changes applied` section to HARDEN.md
After applying : append a `## 10. Changes applied` section to `.claude/audits/HARDEN.md`
with commit-ready summary lines.
Never apply fixes without explicit confirmation. Never use `--no-verify`
@ -563,8 +563,8 @@ URL : <url or static>
Depth : LOCAL | FULL
Mode : audit | fix
Score : XX / 100 (<before> → <after> if fix applied)
Critical alerts : <N> (voir HARDEN.md § 0)
Report : HARDEN.md
Critical alerts : <N> (voir .claude/audits/HARDEN.md § 0)
Report : .claude/audits/HARDEN.md
EXTERNAL VALIDATORS (FULL only) :
Mozilla Observatory : <Grade> (score/135)
@ -613,6 +613,5 @@ NEXT STEPS :
- **SSL Labs can be slow and fail-soft.** 180s poll cap. If TIMEOUT,
note it in HARDEN.md and move on. Cached result auto-hits on next run
via `maxAge=24`. Never block the whole audit waiting on SSL Labs.
- **One report file.** `HARDEN.md` at project root (or `docs/HARDEN.md`
if that convention exists). On re-run, move previous content to a
- **One report file.** `.claude/audits/HARDEN.md`. On re-run, move previous content to a
`## Historique` section, do not overwrite silently.

8
skills/seo/SKILL.md
View File

@ -5,7 +5,7 @@ description: |
Runs the seo-analyzer (classical search — Google, Bing, DuckDuckGo)
and geo-analyzer (AI search — ChatGPT, Perplexity, Claude, Gemini,
Google AI Overviews, Copilot) IN PARALLEL, then consolidates their
output into a unified SEO.md report.
output into a unified .claude/audits/SEO.md report.
Covers: meta, OG, JSON-LD (classical + GEO-optimised schemas),
sitemap, robots.txt (including AI crawlers), llms.txt, headings,
alt attrs, canonicals, hreflang, Core Web Vitals, entity SEO
@ -36,7 +36,7 @@ allowed-tools:
# /seo — parallel SEO + GEO dispatcher
This skill orchestrates TWO specialist agents running in parallel, then
merges their output into a single `SEO.md` report. It is the main
merges their output into a single `.claude/audits/SEO.md` report. It is the main
entry point for any SEO/GEO work on a web project.
## STEP 0 — Collect shared context (ONCE)
@ -267,8 +267,8 @@ the dispatcher will merge your output with seo-analyzer's output.
## STEP 2 — Merge envelopes into SEO.md
Both agents return structured envelopes keyed by SEO.md section
numbers. Consolidate them into a single `SEO.md` at project root
(or `docs/SEO.md` if that convention exists).
numbers. Consolidate them into `.claude/audits/SEO.md`
(run `mkdir -p .claude/audits` first).
### Combined score calculation

25
skills/validate/SKILL.md
View File

@ -4,7 +4,7 @@ description: |
Web standards audit — W3C HTML validity (validator.nu), W3C CSS
validity (jigsaw.w3.org/css-validator), WCAG 2.1 accessibility
(axe-core, pa11y, WAVE API). Dedicated to syntactic and
accessibility conformance. Produces VALIDATE.md at project root.
accessibility conformance. Produces .claude/audits/VALIDATE.md.
Dispatches the validator-analyzer agent with a STRICT scope
filter — no meta/OG/JSON-LD/CWV/security-header noise.
Trigger: "validate", "validation", "w3c", "html validity",
@ -220,7 +220,7 @@ Agent(
"READY TO APPLY — awaiting dispatcher confirmation" at the end
of §5. Do NOT apply any Edit/Write — the dispatcher handles STEP 3.
Output: write <PROJECT_ROOT>/VALIDATE.md per the structure in your
Output: write <PROJECT_ROOT>/.claude/audits/VALIDATE.md (run `mkdir -p .claude/audits` first) per the structure in your
spec (sections 0-8, score XX/100).
"""
)
@ -231,23 +231,23 @@ Agent(
## STEP 2 — Verify output
```bash
test -s VALIDATE.md && wc -l VALIDATE.md || echo "MISSING VALIDATE.md"
test -s .claude/audits/VALIDATE.md && wc -l .claude/audits/VALIDATE.md || echo "MISSING .claude/audits/VALIDATE.md"
```
If missing or empty :
```
⚠️ validator-analyzer did not produce VALIDATE.md. Options :
⚠️ validator-analyzer did not produce .claude/audits/VALIDATE.md. Options :
A) Retry with same scope
B) Downgrade to LOCAL and retry (if FULL failed on network)
C) Abort
```
Extract the score and critical-alert count from VALIDATE.md for the
Extract the score and critical-alert count from `.claude/audits/VALIDATE.md` for the
console summary :
```bash
grep -oE '\*\*Score\*\*\s+:\s+[0-9]+ / 100' VALIDATE.md | head -1
grep -c '^### \[Critique\]' VALIDATE.md
grep -oE '\*\*Score\*\*\s+:\s+[0-9]+ / 100' .claude/audits/VALIDATE.md | head -1
grep -c '^### \[Critique\]' .claude/audits/VALIDATE.md
```
---
@ -256,7 +256,7 @@ grep -c '^### \[Critique\]' VALIDATE.md
Skip this step if `MODE=audit`.
If VALIDATE.md ends with `READY TO APPLY — awaiting dispatcher confirmation` :
If `.claude/audits/VALIDATE.md` ends with `READY TO APPLY — awaiting dispatcher confirmation` :
1. Parse the `## 5. Fix bundle` section.
2. Group by file. For each group, show the combined diff to the user.
@ -276,7 +276,7 @@ Options :
A) Apply all
B) Review each diff before applying
C) Apply only Critique + Haute
D) Abort — keep VALIDATE.md as audit report
D) Abort — keep .claude/audits/VALIDATE.md as audit report
```
4. On `A` : apply each bundle via `Edit` (targeted `old_string` /
@ -284,7 +284,7 @@ Options :
overwriting /seo or /geo content — meta tags, JSON-LD).
5. On `B` : for each diff, show and ask yes/no/skip.
6. On `C` : filter to Critique + Haute, then behave as `A`.
7. On `D` : stop, leave VALIDATE.md untouched.
7. On `D` : stop, leave `.claude/audits/VALIDATE.md` untouched.
After applying, append a `## 8. Changes applied` section with
commit-ready summary lines :
@ -322,7 +322,7 @@ URL : <url or static>
Depth : LOCAL | FULL
Mode : audit | fix
Score : XX / 100 (<before> → <after> if fix applied)
Report : VALIDATE.md
Report : .claude/audits/VALIDATE.md
BREAKDOWN :
W3C HTML : <N errors / M warnings>
@ -367,8 +367,7 @@ Install for better LOCAL coverage :
- **External validators are authoritative on live URLs.** validator.nu
and jigsaw are the W3C backends. If a local tool disagrees with
them, trust the W3C backend; flag the divergence as a finding.
- **One report file.** `VALIDATE.md` at project root (or
`docs/VALIDATE.md` if that convention exists). On re-run, move
- **One report file.** `.claude/audits/VALIDATE.md`. On re-run, move
previous content to a `## Historique` section, do not overwrite
silently.
- **Cache dir.** `.validate-cache/` (gitignored) stores raw tool