The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a local Python script (scripts/l2-state.py) to manage project state. This script executes standard Git commands such as git ls-files and git merge-file to inspect the repository and perform three-way merges of instruction files.- [INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it is designed to ingest and summarize external data from source code, existing documentation, and merged PR reviews.
Ingestion points: SKILL.md phase 1 (code, docs, PR reviews).
Boundary markers: The workflow mandates human approval at every phase (Inspect, Design, Generate, Apply) and explicitly instructs the agent to treat reviews as evidence rather than instructions.
Capability inventory: The scripts/l2-state.py helper can read and write files within the repository and execute local Git commands.
Sanitization: The helper script implements rigorous path validation in the safe_path function to prevent directory traversal and symlink attacks, and it strictly validates the JSON proposal schema.- [DYNAMIC_EXECUTION]: The unit test script (scripts/test_l2_state.py) uses importlib to dynamically load and execute the l2-state.py module for testing purposes, which is a standard development practice.

bento-init