The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill requires the agent to read and interpret Git history, diffs, and branch status, which are entry points for untrusted data. A malicious contributor could craft commit messages or file changes containing instructions designed to influence the agent's behavior during inspection.
Ingestion points: The references/state-and-recovery.md file specifies the use of git log, git diff, git show, and git status to establish the repository state.
Boundary markers: There are no instructions for the agent to use specific delimiters or to ignore potential instructions embedded within the Git output.
Capability inventory: The skill guides the agent in performing sensitive operations including git commit, git push (potentially with force), and git rebase across the local filesystem and remote repositories.
Sanitization: The instructions lack requirements for sanitizing or filtering the content of Git logs or diffs before the agent processes them.

git-operations