The Agent Skills Directory

[COMMAND_EXECUTION]: Uses local shell commands to perform repository analysis and file system tasks. Evidence: git status, git pull, git log, git diff, ls, mkdir, git add, and git commit are executed on local paths in SKILL.md.\n- [DATA_EXFILTRATION]: Accesses git logs and local wiki documents to aggregate retrospective data. Evidence: Reads from ~/wiki/ and local git repositories in SKILL.md. Network usage is limited to standard git pull for synchronization, with no unauthorized data transmission detected.\n- [PROMPT_INJECTION]: The skill processes untrusted data from git logs and wiki files, creating a surface for indirect prompt injection. Evidence: Ingestion points include git log outputs and wiki files (Step 3 in SKILL.md). Capability inventory includes file system writes and git commits (Step 8 in SKILL.md). Sanitization is provided by a structured 'critic' agent with a strict grading rubric (in agents/critic.md) to detect self-rationalization and ensure evidence-based reporting.

retrospective