The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data by reading git diffs, commit messages, and GitHub issues, then interpolating this content into sub-agent prompts.
Ingestion points: Git diff output, commit history, content fetched via gh issue view, and local project files (e.g., docs/, specs/).
Boundary markers: The instructions for sub-agents lack explicit delimiters or escaping instructions for the untrusted data, although they provide clear analytical briefs to guide the sub-agents' focus.
Capability inventory: The skill utilizes git commands, GitHub CLI operations, and file read access.
Sanitization: There is no explicit sanitization of external content before it is passed to the sub-agent prompts.
[COMMAND_EXECUTION]: The skill invokes system commands including git diff, git log, git rev-parse, and gh issue view to perform its analysis.
Evidence: The skill constructs commands using user-supplied inputs, such as git diff <fixed-point>...HEAD.
Validation: The process includes a defensive validation step (git rev-parse <fixed-point>) to verify the integrity of the user-supplied git reference before it is used in subsequent operations, which helps mitigate basic command injection attempts.

code-review