The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses Python's subprocess module and Bash scripting to interact with the Git CLI and system Grep utility. These operations are essential for its primary functions: repository classification, commit message scanning, and diff analysis.
[PERSISTENCE]: The installation script modifies the global Git core.hooksPath configuration to ensure that the security policy engine is invoked for every commit operation on the workstation. This is a documented and standard mechanism for deploying universal Git hooks.
[DYNAMIC_EXECUTION]: The engine supports delegation to repository-local hooks located at .githooks/pre-commit. To mitigate risks, the engine performs its own mandatory security scans before chaining execution and implements a 'delegate-required' verification to ensure the local hook is a regular executable file.
[INDIRECT_PROMPT_INJECTION]: The skill manages an attack surface related to processing user-controlled Git data. 1. Ingestion points: staged diffs, commit messages, and repo-local hook files. 2. Boundary markers: The system uses tiered regex sets (Tier 0 for credentials, Tier 1 for exposure-sensitive data) and global allowlists to define enforcement boundaries. 3. Capability inventory: The skill can block commits, read environment variables, and execute shell scripts. 4. Sanitization: Shell variables generated by the Python sidecar are sanitized using shlex.quote, and delegated hooks undergo file-mode validation.

synthesis-git-hooks