The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted data from Git repositories such as diffs, commit messages, and file contents, which could contain instructions designed to influence the agent's behavior.
Ingestion points: Processes repository data via git worktree list, git diff, and general file system access (SKILL.md).
Boundary markers: The skill does not define specific delimiters to isolate repository data from instructions, relying instead on a structured workflow (SKILL.md).
Capability inventory: Executes shell commands using git and the gh CLI, including actions like git merge, git push, gh pr merge, and file system modifications during conflict resolution (SKILL.md).
Sanitization: The instructions include a specific security check requiring the agent to stop if it detects "suspected secrets" or unrelated changes during preflight inspection (SKILL.md).

merge-worktree