executing-plans

Pass

Audited by Gen Agent Trust Hub on Mar 2, 2026

Risk Level: SAFEPROMPT_INJECTIONNO_CODE
Full Analysis
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it executes instructions sourced from external implementation plans.\n
  • Ingestion points: Plan data is ingested from a file in 'Step 1: Load and Review Plan'.\n
  • Boundary markers: The skill does not define or use explicit delimiters or markers to isolate external plan content from the agent's system instructions.\n
  • Capability inventory: The workflow involves executing tasks and running verifications, which frequently requires shell command execution or direct code modification capabilities.\n
  • Sanitization: The skill relies on a manual 'critical review' by the agent rather than automated input sanitization or validation routines.\n- [NO_CODE]: The skill consists entirely of instructional Markdown content and does not bundle any scripts, binaries, or automated installation logic.\n- [SAFE]: No evidence of hardcoded credentials, data exfiltration, or obfuscation was found. The skill includes explicit safety instructions, such as prohibiting implementation on the main branch without user consent.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 2, 2026, 04:11 AM