rules-capture

Pass

Audited by Gen Agent Trust Hub on Apr 22, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill captures untrusted user input from conversations and writes it to governance files located in .agents/SYSTEM/. This functionality introduces a surface for indirect prompt injection where malicious instructions could be codified as persistent system rules. The skill's design includes a human-in-the-loop confirmation step to mitigate this risk.\n
  • Ingestion points: User conversation text is monitored and extracted via regex triggers defined in SKILL.md.\n
  • Boundary markers: Extracted user quotes are encapsulated in markdown blockquotes (>) to provide structural separation.\n
  • Capability inventory: The skill performs file append, move, and removal operations targeting files within the .agents/SYSTEM/ directory as described in SKILL.md.\n
  • Sanitization: The skill does not perform automated sanitization of user input, instead relying on documentation formatting and a mandatory user approval step before rules are finalized.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 22, 2026, 04:25 AM