The Agent Skills Directory

[SAFE]: The skill implements strong security-positive architectural patterns. It explicitly prohibits autonomous execution by requiring human sign-off at every 'stage gate' and enforces the use of external 'oracles' (e.g., bibtex validators, LaTeX compilers, and DOI resolvers) for verification. This design effectively mitigates risks associated with hallucinations and indirect prompt injection by ensuring that the AI cannot validate its own output or potentially malicious external content without objective, measurable signals.
[SAFE]: The included Python script scripts/pipeline_state.py adheres to security best practices for agent skills. It utilizes only the Python standard library, performs no network operations, and restricts file system interactions to the local workspace using safe serialization methods (JSON).

orchestrate-paper