The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes uv run python scripts/pipeline.py and scripts/run.py to update decision ledgers and manage pipeline state. This is a standard and expected pattern for repository-local tooling.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and review external data, including UNITS.csv, STATUS.md, and various pipeline Artifacts. While this presents a theoretical attack surface where malicious content in artifacts could attempt to influence the agent, the risk is mitigated by the skill's primary purpose of facilitating human review and the lack of automated downstream execution of artifact content.
Ingestion points: UNITS.csv, STATUS.md, DECISIONS.md, and pipeline Artifacts (file-based data).
Boundary markers: None explicitly defined in the instructions for separating artifact content from instructions.
Capability inventory: File system writes to DECISIONS.md and execution of local Python scripts via the uv tool.
Sanitization: Not present in the provided instruction set or helper script; the skill relies on the agent's inherent review capabilities and the human-in-the-loop requirement.

human-checkpoint