| name | scan-harness |
| description | Audit the user's Claude Code agent harness for improvements — global config, project config, recent session patterns, and opportunities for better guardrails and feedback loops. Use when the user asks to improve their Claude Code setup. |
| model | opus |
| effort | high |
Context
Use up to 25 subagents to explore the following data sources:
- Study the current project's documented agent harness intentions (if any)
- Study the user's global Claude Code configuration in
~/.claude/ (custom skills, agents and settings are symlinked from ~/Repos/ooloth/dotfiles/tools/claude/config/)
- Study the project-level Claude Code configuration in
./.claude/
- Study the user's recent Claude Code sessions in
~/.claude/projects/
- Study the Claude Code skills docs: https://code.claude.com/docs/en/skills
- Study the Claude Code hooks docs: https://code.claude.com/docs/en/hooks-guide
- Study the Claude Code subagents docs: https://code.claude.com/docs/en/sub-agents
- Study the Claude Code agent teams docs: https://code.claude.com/docs/en/agent-teams
- Study the Claude Code /loop docs: https://code.claude.com/docs/en/scheduled-tasks
Your task
- Identify opportunities to improve the user's global Claude Code configuration
- Identify opportunities to improve the current repo's project-level Claude Code configuration
- Identify opportunities in the patterns you see across recent Claude Code sessions
- Look for opportunities that would provide the context needed to help agents understand the intent of the project and make correct judgement calls
- Look for opportunities to introduce clear guardrails helping agents validate their work
- Focus especially on opportunities to introduce deterministic mechanisms that introduce reliable back pressure and a feedback loop that makes it obvious to agents what choices are correct and what invariants they must uphold at all times
- Rank the findings by priority for the user to address (based on impact, cost of delay, ROI, etc)
- Present the prioritized findings to the user, ensuring a summary table is included near the end of your response for clarity
- Generate a self-contained HTML slide deck of the prioritized findings:
mkdir -p .outputs/<yyyy-mm-dd>- Write the report to
.outputs/<yyyy-mm-dd>/scan-harness.html — clean minimal styling, one slide per category plus a summary/title slide, keyboard arrow-key and click navigation between slides
open .outputs/<yyyy-mm-dd>/scan-harness.html
- Recommend a next action and wait for the user's response
