The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on the execution of local bash scripts (scripts/prepare-target-branch.sh and scripts/sync-target-branch.sh) and standard CLI tools (git, gh). These tools are used for repository management tasks such as rebasing, pushing, merging, and worktree cleanup. The instructions emphasize that the agent should only perform pushes, merges, or cleanup when explicitly authorized by the user.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external command outputs, including git status, git diff, and gh pr view, which are potential ingestion points for indirect instructions embedded in repository content or PR metadata.
Ingestion points: The agent is instructed to read staged diffs and PR states from the repository environment.
Boundary markers: The skill includes explicit naming and commit subject guardrails in SKILL.md, instructing the agent to suggest corrections or ask for confirmation if inputs are vague or non-conventional.
Capability inventory: The skill possesses the capability to modify the local repository state via rebase, push changes, and remove branches or worktrees.
Sanitization: The skill relies on the agent's adherence to instructions to verify that staged changes match their descriptions, rather than using automated sanitization of command output.
[DYNAMIC_EXECUTION]: The validate.sh script uses a shell heredoc to execute a dynamically generated Python script for structural verification. This is limited to local validation of the skill's integrity and is not used during runtime agent execution.

git-branching