To improve an agent:
- Test the agent on real codebases
- Identify false positives or missed issues
- Update the agent's detection patterns
- Submit a PR with examples
Each agent follows this pattern:
---
name: agent-name
description: What this agent checks
tools: "*"
---
## DESCRIPTION
What the agent does and when to use it.
## OUTPUT FORMAT
🔍 **[filename]:[line]** • [CRITICAL/HIGH/MEDIUM]
**Problem**: [Issue description]
**Fix**: [Concrete solution]Test your changes by:
- Copy modified agent to a test project's
.claude/agents/ - Run via Claude Code Task tool
- Verify the agent works correctly
- Check that reports are actionable