The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a bundled script (scripts/review-state.mjs) to perform Git operations such as git rev-parse, git status, and git merge-base to track review progress and branch lineage.- [REMOTE_CODE_EXECUTION]: The instructions in SKILL.md include shell commands to install missing dependencies using pnpm dlx from external sources. This pattern downloads and executes remote code at runtime, posing a risk if the agent attempts to automate environment setup.- [INDIRECT_PROMPT_INJECTION]: The skill ingests un-sanitized data from Git diffs and project issues to drive automated code reviews and fixes.
Ingestion points: Untrusted data enters the context via git diff output and the body of issues resolved in Step 0.
Boundary markers: No specific delimiters or safety instructions are defined to prevent the agent from obeying instructions embedded within the code changes or issue text.
Capability inventory: The skill has the ability to execute Git commands, write state files, and trigger automated code modifications through the implement skill.
Sanitization: The skill lacks explicit filtering or escaping of external content before passing it to sub-agents.

polish