The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted data from the repository's git history which could be manipulated to influence agent behavior. * Ingestion points: The agent reads git log and git diff outputs in the first step of the workflow defined in SKILL.md. * Boundary markers: There are no explicit delimiters or instructions provided to the agent to treat content from the git history as untrusted or to ignore embedded instructions. * Capability inventory: The skill has the ability to execute shell commands during the history rebuild and validation phases in REBUILD.md. * Sanitization: There is no evidence of sanitization or validation of the content read from the git history before it is used to design the commit sequence and validation plan.
[COMMAND_EXECUTION]: The skill relies on extensive execution of shell and git commands to manipulate the repository state. * It uses complex git operations including git rebase -i --autosquash, git apply with binary patches, and git worktree management. * It references an external tool git-hunk for granular staging of changes.
[DYNAMIC_EXECUTION]: The validation process executes shell commands that are dynamically generated by the agent during the planning phase. * Evidence: In REBUILD.md, the validation loop executes <checks planned for this commit> in a temporary worktree. These checks are strings generated by the agent in SKILL.md Step 1.

recommit