The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes Python scripts (agents_block.py, push_gate.py) to execute local git and gh commands via subprocess.run. These operations are used to inspect repository state, verify commit history, and identify the top-level directory of a project.
[PERSISTENCE]: Installs a persistent PreToolUse hook in the agent's configuration files (settings.json or hooks.json). This hook intercepts shell commands to enforce a review gate policy across agent sessions.
[INDIRECT_PROMPT_INJECTION]: The skill establishes an ingestion surface by parsing shell commands and local instruction files to determine the application of review policies.
Ingestion points: push_gate.py reads tool-call JSON from stdin and reads instruction overlays like .agents/bento/bento-ship.md.
Boundary markers: Employs marker-bounded blocks (e.g., ) to manage injected content within AGENTS.md.
Capability inventory: Executes git and gh CLI tools for repository verification.
Sanitization: Uses shlex.split and shlex.quote to safely handle command tokens and script paths.

bento-setup