The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs project setup and hook validation using commands such as prek install, prek run --all-files, and system package manager commands (cargo install, brew install). These actions are standard for the tool's intended use and are gated by mandatory user authorization steps defined in the protocol.
[EXTERNAL_DOWNLOADS]: The skill retrieves hook definitions and version information from external sources on GitHub, including repositories like astral-sh/ruff-pre-commit and shellcheck-py/shellcheck-py. These are widely used and reputable sources in the developer community.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data by reading local project configuration files (e.g., package.json, Cargo.toml, prek.toml) to provide tailored hook suggestions. While this reading of project files creates a potential attack surface, the risk is mitigated by the skill's requirement to present a diff of all proposed changes to the user and obtain explicit permission before execution.
Ingestion points: The agent reads repository root files including Cargo.toml, pyproject.toml, setup.py, requirements.txt, package.json, go.mod, Gemfile, and prek.toml.
Boundary markers: None explicitly mentioned for the file-reading process; it relies on the agent's standard file-system interface.
Capability inventory: Ability to execute shell commands via prek, cargo, brew, npx, gofmt, and pytest depending on the project type.
Sanitization: The skill incorporates a mandatory human-in-the-loop step, requiring the agent to show a diff and ask the user before running any mutating or installation commands.

prek