The Agent Skills Directory

[SAFE]: The skill is designed for code analysis and refactoring, using standard tools like git diff and specialized MCP tools for structural analysis. The workflow is transparent, involving triage, parallel review agents, and a fix phase. No malicious patterns such as data exfiltration, credential harvesting, or unauthorized remote code execution were detected.
[PROMPT_INJECTION]: The skill processes untrusted code changes, which presents an indirect prompt injection surface.
1. Ingestion points: git diff output and code-review-graph tool outputs (SKILL.md).
2. Boundary markers: No explicit delimiters or instructions are used to isolate ingested code from the agent's instructions.
3. Capability inventory: The skill can modify repository files (Phase 3) and execute shell commands (git).
4. Sanitization: No explicit sanitization of the ingested code is implemented. This surface is inherent to code-processing tools and is assessed as safe in the context of a development assistant.

simplify