The Agent Skills Directory

[SAFE]: The skill is designed to manage internal project documentation and does not exhibit malicious patterns. It adheres to safety principles such as mandatory human-in-the-loop review and a strict separation between analysis and file-writing phases.- [COMMAND_EXECUTION]: The skill utilizes Git commands (git log, git merge-base, git commit) for historical analysis and record-keeping. The skill instructions in SKILL.md contain explicit safety rules that prohibit destructive Git operations such as git reset, git rebase, and git push --force.- [PROMPT_INJECTION]: The skill ingests untrusted data from project artifacts and commit messages, which presents a surface for indirect prompt injection. This risk is managed through a multi-layered evidence chain: 1. Ingestion points: Reads from docs/cr/*.md and Git logs. 2. Boundary markers: Relies on human-in-the-loop review rather than automated delimiters. 3. Capability inventory: Limited to Git analysis and documentation updates. 4. Sanitization: Relies on mandatory per-tier human approval to filter out malicious or irrelevant suggestions before they are written to the project's instructions.

checkpoint-distill