Skills
Navigate
skills/romainsimon/skills-for-decision-making/learning-from-outcomes/Gen Agent Trust Hub

learning-from-outcomes

Pass

Audited by Gen Agent Trust Hub on Sep 23, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes a local script using node scripts/calc.js. This script processes input files predictions.json and result.json to calculate calibration metrics and credit assignment.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external JSON files and incorporates the results into the agent's output template.\n
  • Ingestion points: Data is ingested from predictions.json and result.json as specified in the SKILL.md workflow.\n
  • Boundary markers: The instructions lack explicit delimiters or warnings to ignore instructions within the data files.\n
  • Capability inventory: The skill environment allows command execution (node) and file access to support calculation scripts.\n
  • Sanitization: There is no evidence of data sanitization or integrity checks for the ingested JSON files.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 23, 2026, 12:08 PM