claude-improve-config
Skill126 starsChanged 9 months agoApache-2.0View on GitHub
Self-reflect on the current session to identify mistakes and propose improvements to .claude configuration (CLAUDE.md, hooks, skills).
In this repository7 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- docs-page-frontmatterSkill
- hypothesisSkill
- narwhalsSkill
- sybilSkill
- syrupySkill
- tachSkill
The file
We split this skill into the rules it gives an agent
12 rules. About 0 are common, 11 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈11
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 12
After a session with significant mistakes or rework
- Distinctive: line 13
When you notice patterns that could be prevented with better configuration
- Very specific: line 14
When explicitly requested via /claude-improve-config
- Distinctive: line 38
Only propose changes for critical or moderate issues.
- Distinctive: line 54
Stop hooks to enforce checks before completion
- Distinctive: line 89
Only propose changes that are generalizable - don't add rules for one-off situations
- Distinctive: line 90
Keep CLAUDE.md concise - prefer hooks for enforcement over verbose instructions
- Distinctive: line 91
Test proposed hooks mentally - ensure they won't block legitimate workflows
- Distinctive: line 92
Prefer minimal changes - one well-designed rule is better than many narrow ones
- Distinctive: line 93
Consider false positives - hooks should not create friction for normal operations
- Distinctive: line 101
Proposal: Add to CLAUDE.md:
- Distinctive: line 111
Proposal: Add PostToolUse hook for Write/Edit:
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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
Claude Config Self-Improvement
Analyze the current session for errors, mistakes, or inefficiencies and propose improvements to the .claude configuration to prevent similar issues in the future.
When to Use
- After a session with significant mistakes or rework
- When you notice patterns that could be prevented with better configuration
- When explicitly requested via
/claude-improve-config - Automatically triggered by the
SessionEndhook
Self-Reflection Process
1. Analyze the Session
Review the conversation for:
- Repeated mistakes: Same error made multiple times
- Misunderstandings: User had to correct Claude's interpretation
- Inefficient workflows: Better approaches discovered late in the session
- Tool misuse: Wrong tool for the job, excessive tool calls
- Missing context: Information that would have helped if in CLAUDE.md
- Ignored instructions: Existing CLAUDE.md rules that were violated
2. Categorize Issues
Rate severity:
- Critical: Caused significant rework or user frustration
- Moderate: Caused delays or minor corrections needed
- Minor: Slight inefficiency but didn't impact outcome
Only propose changes for critical or moderate issues.
3. Propose Configuration Changes
Changes can include:
CLAUDE.md Updates:
- New guardrails or constraints
- Clarified existing instructions
- Project-specific patterns to follow
New Hooks:
- PreToolUse hooks to prevent specific mistakes
- PostToolUse hooks to validate outputs
- Stop hooks to enforce checks before completion
New Skills:
- Reusable knowledge for recurring tasks
- Best practices for specific domains
4. Format the Proposal
Present proposals clearly:
## Session Reflection
### Issues Identified
1. **[Critical/Moderate]** Brief description of the issue
- What happened: ...
- Why it happened: ...
- Impact: ...
### Proposed Changes
#### Change 1: [Type - CLAUDE.md/Hook/Skill]
**Rationale:** Why this change prevents the issue
**Implementation:**
[Show the exact changes to make]
#### Change 2: ...Guidelines
- Only propose changes that are generalizable - don't add rules for one-off situations
- Keep CLAUDE.md concise - prefer hooks for enforcement over verbose instructions
- Test proposed hooks mentally - ensure they won't block legitimate workflows
- Prefer minimal changes - one well-designed rule is better than many narrow ones
- Consider false positives - hooks should not create friction for normal operations
Example Proposals
Example 1: Missing Test Verification
Issue: Claude claimed tests passed without running them.
Proposal: Add to CLAUDE.md:
**Test verification**: Never claim tests pass without showing actual test output. Always run `uv run pytest` with the specific test file before marking test-related tasks complete.Example 2: Repeated Lint Failures
Issue: Code was submitted with lint errors multiple times.
Proposal: Add PostToolUse hook for Write/Edit:
{
"matcher": "Write|Edit",
"hooks": [
{
"type": "command",
"command": "ruff check --select=E,F $TOOL_INPUT.file_path 2>/dev/null || true"
}
]
}Example 3: Wrong Branch for PR
Issue: PR was created against wrong base branch.
Proposal: Add to CLAUDE.md:
**Pull requests**: Always verify the target branch before creating a PR. For this project, PRs should target `main` unless explicitly specified otherwise.Non-Examples (Do NOT Propose)
- One-off mistakes that won't recur
- User preference differences (not errors)
- Issues already covered by existing configuration
- Overly specific rules that won't generalize
- Changes that would slow down normal workflows significantly
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.

