The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from Git repositories (commit messages, diffs, branch names) which could contain malicious instructions designed to influence the agent's behavior. Ingestion points: Outputs from commands like git log, git diff, and git show (found in SKILL.md, workflows/Log.md, workflows/Diff.md). Boundary markers: Absent; there are no specific instructions or delimiters used when the agent is prompted to interpret or summarize repository history. Capability inventory: The skill leverages the Git CLI for significant operations including git push (remote data modification), git reset --hard (destructive local state changes), and git bisect run (execution of arbitrary local scripts). Sanitization: While the skill proactively checks for hardcoded secrets and sensitive file types, it does not implement sanitization to neutralize natural language instructions embedded within commit messages.
[COMMAND_EXECUTION]: The skill relies extensively on executing shell commands via the Git CLI to perform its core functions. Evidence: Found across all workflow files utilizing commands like git rebase, git cherry-pick, git bisect, and git worktree.

git