The Agent Skills Directory

[SAFE]: The skill operates within the repository context using allowed tools to read code history and write documentation files. It follows a well-defined process that prioritizes evidence-based drafting and restricts the agent from making claims not present in the code history.
[COMMAND_EXECUTION]: The skill utilizes git commands such as git diff, git log, and git status to retrieve the change history necessary for drafting documents. These commands are used legitimately to gather repository state and do not involve unsanitized user input in a way that would lead to command injection.
[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted data from git commit messages and diffs. Evidence chain: 1. Ingestion points: output from git logs, git diffs, and project plan files in the projects/ directory. 2. Boundary markers: The instructions contain multiple strict constraints requiring the agent to trace every claim to a specific hunk or commit and explicitly forbidding the use of rationale not found in the source. 3. Capability inventory: The skill has the ability to write to files like CHANGELOG.md and docs/ using the Write and Edit tools. 4. Sanitization: No explicit data sanitization is performed. The risk is considered safe because the output is human-facing documentation and the instructions emphasize strict adherence to provided evidence.

document