The Agent Skills Directory

[PROMPT_INJECTION]: The skill is subject to indirect prompt injection as it ingests untrusted data from the repository metadata.
Ingestion points: Data is read from branch names, commit messages, and pull request titles via git and gh commands in SKILL.md.
Boundary markers: The skill does not define boundary markers to isolate untrusted repository data from instructions.
Capability inventory: The skill can perform potentially destructive operations including git branch -d, git remote prune, and git stash drop.
Sanitization: There is no evidence of sanitization or filtering applied to the data retrieved from the repository commands.
[COMMAND_EXECUTION]: The skill relies on executing shell commands to audit and modify the Git repository state.
Evidence: Commands include git status, git stash, git fetch, git worktree, and gh pr list.
Safety: The skill includes explicit guardrails requiring user confirmation for destructive actions like deleting non-merged branches or closing PRs.

repository-cleanup