The Agent Skills Directory

Local Command Execution: The skill executes a local Python script (scripts/criticality_score.py) to process maintenance data. This is a common pattern for ensuring that calculations remain consistent and do not rely on probabilistic model outputs. The execution is limited to the provided script using standard system libraries.\n- Indirect Prompt Injection Surface: The skill processes data from external files, specifically ranked-causes.json, to determine priority levels. This represents a potential area for review as untrusted data entering the agent context can influence behavior.\n
Ingestion points: Data enters the skill's logic through the ranked-causes.json file as specified in the execution steps of SKILL.md.\n
Boundary markers: The skill includes explicit instructions for the agent to quote the engine output verbatim and forbids the model from setting priorities independently, which serves as a boundary against instructions embedded in data.\n
Capability inventory: The skill utilizes the Python script to read from and write to the local file system on paths provided via command-line arguments.\n
Sanitization: The script parses data into structured JSON and validates inputs against pre-defined failure modes and criticality classes defined in the reference documents, reducing the likelihood of unexpected input handling.

criticality-rank