review-claude-config
Skill0 starsChanged 4 months agoMITView on GitHub
Audit and optimize agent config files — a single CLAUDE.md or AGENTS.md, or the whole ecosystem of global and per-repo CLAUDE.md, CLAUDE.local.md, memory files, and skills — for redundancy, inconsistency, outdated facts, and misplaced content. Use when asked to "review CLAUDE.md", "optimize CLAUDE.md/AGENTS.md", "audit claude config", "clean up claude files", or "check what should be global vs local".
In this repository12 other files it gives its agents.
- englogSkill
- githubSkill
- git-repos-cleanupSkill
- gitSkill
- golang-developmentSkill
- pull-requestSkill
- python-developmentSkill
- review-claude-settingsSkill
- review-pull-requestSkill
- session-learningsSkill
- triage-dependency-prsSkill
- untracked-secrets-scanSkill
The file
We split this skill into the rules it gives an agent
6 rules. About 0 are common, 6 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈6
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 29
Run these in parallel:
- Distinctive: line 129
If no findings in a category, state "No findings."
- Distinctive: line 135
After presenting the report, ask: "Apply all?
- Distinctive: line 150
Never rewrite an entire file unless the user explicitly asks.
- Distinctive: line 151
Never silently resolve cross-repo inconsistencies — always present both versions and ask which wins.
- Distinctive: line 152
Do not promote repo-specific content to global just because it appears in multiple repos.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Review Claude Config
Audits and optimizes agent config files and produces a structured findings report. Changes require explicit confirmation before applying.
Two modes:
- Single file (user names one
CLAUDE.md/AGENTS.md): run Phase 2 and the Optimization checks on that file only, then Report and Apply. - Ecosystem audit (default): run all phases.
Scope
- Global:
~/.claude/CLAUDE.md - Per-repo:
CLAUDE.mdandCLAUDE.local.mdunder the user-provided workspace root (all depths) - Memory:
~/.claude/projects/<project>/memory/(index + files) - Skills:
~/.claude/skills/*/SKILL.md
Phase 1: Discovery
Run these in parallel:
find <workspace-root> -maxdepth 4 \( -name "CLAUDE.md" -o -name "CLAUDE.local.md" \) 2>/dev/null | sortls ~/.claude/skills/Then read in parallel:
~/.claude/CLAUDE.md(global config)~/.claude/projects/<project>/memory/MEMORY.md(memory index, if it exists)- Each discovered repo
CLAUDE.mdandCLAUDE.local.md
Phase 2: Audit Global CLAUDE.md
Check for:
Redundancy — same fact stated twice across sections. Decision: keep the more specific instance, remove or collapse the other.
Internal inconsistency — conflicting facts within the file (e.g. one section says "use X" and another says "use Y for the same case").
Outdated facts — tools, workflows, or systems that no longer apply. Red flags: tool names not in the active Tooling table, processes that contradict the CI/Workflow section, dates that have passed.
Bloat — content generic enough to belong in a skill file, or specific enough to belong in a per-repo CLAUDE.local.md. Global rules should apply across all repos. Repo-specific conventions (commit format, nav file, link syntax) belong locally.
Missing facts from memory/skills — if a memory file or skill contains a rule the user would expect the global CLAUDE.md to enforce, flag it for promotion.
Phase 3: Audit Per-Repo Files
For each repo's CLAUDE.md / CLAUDE.local.md:
Should be global — a rule applies to all repos, not just this one (e.g. a git convention, a tool flag, a security rule). Flag for promotion to ~/.claude/CLAUDE.md.
Duplicates global — a rule already covered verbatim by the global file. Flag for removal from the repo file (avoid drift).
Outdated — references to tools, branches, or processes that no longer exist in this repo.
Cross-repo inconsistency — two repos have conflicting guidance for the same concept (e.g. different commit message formats). Flag; do not silently resolve.
Phase 4: Memory and Skills Cross-Check
Memory → Global CLAUDE.md: A memory file records a fact or user preference that should be a standing rule (not ephemeral state). Flag for promotion. Criteria: would the user expect this to apply in every future session, not just the one where it was learned?
Skills → Global CLAUDE.md: A skill documents a tool-specific workflow detail that should instead be a global policy (e.g. "always confirm before posting to GitHub"). Conversely, if global CLAUDE.md restates what a skill already covers fully, the global entry is redundant.
Optimization Checks (any file)
Inconsistency detection — beyond outright contradictions, look for: the same concept under different names, two workflows describing the same process with different steps or order, conflicting naming or directory conventions, and mixed formatting for the same kind of content. Resolve to the most specific, most recent, or most referenced version; when both are valid in different contexts, keep both with the condition stated. Never pick silently when the authoritative version is unclear: ask.
What to move where:
- Detailed step-by-step workflow already in a skill or instruction file → replace with a 1-3 line summary + reference.
- Rule that applies to every repo → global file; repo-specific convention → that repo's file.
- Personal or machine-specific detail in a team-shared file →
CLAUDE.local.md. - Human-only notes, history, evolution notes, generic advice the model already knows → remove.
- Several examples of one concept → keep one, genericized.
Size reduction is a byproduct, not the goal.
Phase 5: Report
Present findings as a structured table before any changes. Group by severity:
## Findings
### Global CLAUDE.md
| # | Category | Section | Issue | Proposed fix |
|---|---------------|-----------------|-------------------------------------|---------------------------|
| 1 | Redundancy | Tooling / Git | GPG signing stated twice | Remove from Tooling table |
| 2 | Inconsistency | CI / Tooling | gh described as CI tool | Update gh description |
| 3 | Outdated | CI / Workflow | Deprecated process still mentioned | Update the workflow guidance |
### Per-Repo Files
| # | File | Category | Issue | Proposed fix |
|---|--------------------------|-----------------|-------------------------------------|---------------------------|
| 1 | repo/CLAUDE.md | Should be global | Commit format rule | Move to global CLAUDE.md |
### Memory / Skills
| # | Source | Issue | Proposed fix |
|---|--------------------------|-------------------------------------|---------------------------|
| 1 | memory/feedback_*.md | Preference not reflected globally | Add rule to global CLAUDE.md |If no findings in a category, state "No findings."
Phase 6: Apply (with confirmation)
After presenting the report, ask: "Apply all? Or specify which numbers to apply."
- Apply only the confirmed items.
- For global CLAUDE.md edits: use targeted
Editcalls (one per finding); do not rewrite entire sections. - For repo file edits: same — targeted edits only.
- For memory/skill promotions: add the rule to the appropriate section in global CLAUDE.md; do not delete the source unless it is truly redundant.
- Re-read each edited section after applying to confirm the change landed correctly.
- Verify preservation: diff the file before/after and check every removed actionable rule (commands, flags, paths, constraints, checklists) either still exists elsewhere in the file, moved to a referenced location, or was an approved removal. Report anything else as lost and restore it.
Constraints
- Never rewrite an entire file unless the user explicitly asks.
- Never silently resolve cross-repo inconsistencies — always present both versions and ask which wins.
- Do not promote repo-specific content to global just because it appears in multiple repos; check whether it is genuinely universal first.
- Line length ≤120 characters in CLAUDE.md files.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

