thrifty-check

Pass

Audited by Gen Agent Trust Hub on Jun 30, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection by processing external files and system outputs.
  • Ingestion points: The skill reads "/briefs/.md", "CONTRACT.md", and the output of failing tests.
  • Boundary markers: No boundary markers or specific instructions to treat external data as untrusted are present in SKILL.md.
  • Capability inventory: The agent has the ability to execute shell commands ("Re-run the failing command") and modify files ("Make the minimal in-place edits").
  • Sanitization: No evidence of sanitization or filtering of the ingested data before it influences agent behavior or command execution.
  • [COMMAND_EXECUTION]: The skill is designed to run arbitrary shell commands based on external test failures.
  • Evidence: The instructions state "Re-run the failing command yourself to confirm" and "Re-run all of the unit's criteria".
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 30, 2026, 12:23 AM
Security Audit — agent-trust-hub — thrifty-check