audit-agent-history — Claude Code skill: what it does, how to install
Scans local AI agent transcripts to isolate recurring mistakes and turn them into project safety rules. Helpful when an assistant repeats the same errors across sessions.
npx skills add timschoch/skilly --skill audit-agent-history -a claude-codeWhat audit-agent-history does
The skill parses session histories located in ~/.claude/projects/ and ~/.codex/sessions/. It scans for phrases where the user countermands an assistant action, tracking triggers like "no", "stop", or "revert". Corrections are grouped by failure category and calculated relative to message counts, normalizing rates across different models.
Processing runs inside a single dedicated subagent. This keeps heavy raw transcripts out of your primary context window. The skill evaluates the worst session with an identical model to locate why rules failed, then inspects the slowest session to identify tool calls that altered no output.
The final output suggests up to three actionable rules starting with Never, complete with authentic positive and negative snippets. Without pre-existing JSONL logs in the default paths, the skill has nothing to evaluate. Approved hazard rules sync directly to the central hub rather than local repository overrides.
What it helps with
- inspect session histories to identify recurring agent mistakes
- convert frequent failure patterns into strict workflow rules
- isolate unhelpful tool invocations in bloated transcripts
- hoist shared local guardrails into the upstream rule repository
Example prompts for the agent
- Audit my Claude Code session logs from the past 30 days and extract common failure modes
- Analyze recent Codex transcripts to draft hazard rules for recurring mistakes
- Inspect the longest agent sessions and flag tool calls that produced no changes
How to install audit-agent-history
1Skills CLI (Vercel)
npx skills add timschoch/skilly --skill audit-agent-history -a claude-code2Manually: put the files in place
~/.claude/skills/audit-agent-history/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/audit-agent-history/
1Skills CLI (Vercel)
npx skills add timschoch/skilly --skill audit-agent-history -a codex2Manually: put the files in place
.agents/skills/audit-agent-history/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add timschoch/skilly --skill audit-agent-history -a cursor2Manually: put the files in place
.cursor/skills/audit-agent-history/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install timschoch/skilly audit-agent-history2Manually: put the files in place
.github/skills/audit-agent-history/SKILL.md1Skills CLI (Vercel)
npx skills add timschoch/skilly --skill audit-agent-history -a gemini-cli2Platform CLI
gemini skills install https://github.com/timschoch/skilly --path bundles/workflow/skills/audit-agent-historyManually: put the files in place
.devin/skills/audit-agent-history/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add timschoch/skilly --skill audit-agent-history -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about audit-agent-history
How to install audit-agent-history in Claude Code?
Run npx skills add timschoch/skilly --skill audit-agent-history -a claude-code. You can also place the skill file directly at ~/.claude/skills/audit-agent-history/SKILL.md.
What does audit-agent-history do?
It mines local session logs to detect when users undo or correct assistant actions. The skill quantifies these failure modes and generates targeted guardrails based on real workspace history.
Why does the audit run inside a subagent?
Parsing multiple raw session logs consumes significant context. Running the analysis in an isolated subagent keeps your main conversation clean, returning only the summarized findings.
