The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates local Git management through a wide range of commands, including those for history modification such as rebase, reset, and filter-repo. Safety is maintained by instructions requiring recovery refs and explicit authorization for state-changing operations.\n- [DYNAMIC_EXECUTION]: The skill provides instructions to configure the local Git environment to run a hook script from its assets (assets/commit-msg). It also includes templates for setup and audit scripts which are executed locally.\n- [EXTERNAL_DOWNLOADS]: Reference documentation includes examples of fetching repository patches and diffs from GitHub using curl. These operations target a well-known, trusted service.\n- [REMOTE_CODE_EXECUTION]: The documentation describes piping external patches directly into Git commands (curl ... | git am). While this executes remote content, the source is a trusted service (GitHub).\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from Git logs, diffs, and external patches, creating an attack surface for indirect prompt injection.\n
Ingestion points: Git history, diff outputs, repository status, and downloaded patch files.\n
Boundary markers: The 'commit-slicing contract' and validation steps like git diff --check serve as delimiters and check points.\n
Capability inventory: Git command execution, hook configuration, and local file writes.\n
Sanitization: The commit-msg hook uses regex to validate inputs.\n- [PRIVILEGE_ESCALATION]: The skill instructs the user to apply chmod +x to grant necessary execution permissions to hook scripts.

git-workflows