skill-supervisor

Pass

Audited by Gen Agent Trust Hub on Aug 20, 2026

Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructs the agent to execute a local evaluation script using python3 evals/pm/evaluate.py --decisions results/decisions_vN.jsonl --judge results/judgments_vN.jsonl. This is part of the intended evaluation harness functionality.
  • [PROMPT_INJECTION]: The skill is designed to ingest and iterate on external data (market scenarios and candidate skills), creating an indirect prompt injection surface. This is a characteristic of its primary purpose as an optimization tool.
  • Ingestion points: Market scenarios from evals/pm/scenarios/inputs.jsonl and candidate SKILL.md files generated by the Modifier subagent.
  • Boundary markers: The workflow uses role separation (blind executors) to mitigate influence, though no explicit string-based delimiters are defined for the input data files.
  • Capability inventory: The skill can execute shell commands (python3), perform file operations in the results/ directory, and spawn subagents with broad tool access.
  • Sanitization: No explicit sanitization or validation logic is specified for the external JSONL scenarios or the dynamically generated markdown files before processing.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 20, 2026, 02:21 PM
Security Audit — agent-trust-hub — skill-supervisor