shh
Pass
Audited by Gen Agent Trust Hub on Sep 2, 2026
Risk Level: SAFEPROMPT_INJECTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill employs session-wide persistence mechanisms and override instructions ("These rules apply to every response for the rest of the session", "Turn them off only when...", "If you are unsure whether they still apply, they do.") that are typically used to maintain specific behaviors across conversation turns.
- [PROMPT_INJECTION]: The directives include explicit concealment requirements ("Never name or announce Shh"), which is a pattern used to hide instruction-based overrides from the user.
- [INDIRECT_PROMPT_INJECTION]: The skill creates an attack surface by instructing the agent to process and quote potentially untrusted external data like error logs and documentation without sanitization or boundary markers. 1. Ingestion points: Error logs and source documentation mentioned in SKILL.md instructions. 2. Boundary markers: Absent; instructions require exact quoting. 3. Capability inventory: Documentation mentions shell execution (npm, git, rm) and file modification. 4. Sanitization: None specified; instructions demand preserving original content.
Audit Metadata