clarify-decisions

Pass

Audited by Gen Agent Trust Hub on Sep 5, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONNO_CODE
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is highly dependent on user input to navigate a complex decision tree and recompute the 'open set' of questions. This input is processed without explicit sanitization or boundary markers, creating an attack surface where a user could potentially steer the agent's logic through embedded instructions.
  • Ingestion points: The skill explicitly mandates that 'EVERY QUESTION IS INLINE CHAT' and the agent must 'recompute the open set after every answer' (SKILL.md).
  • Boundary markers: There are no requirements in the instructions to wrap user responses in delimiters or to ignore instructions within the data.
  • Capability inventory: The agent has the capability to read project metadata files (Knowns, scan digests) and invoke sub-skills like load-subgraph and assess-observability (SKILL.md).
  • Sanitization: No sanitization or validation of user-provided content is described in the protocol.
  • [NO_CODE]: The skill consists entirely of markdown instructions, evaluation data, and configuration metadata. It does not include or execute any script files such as Python, JavaScript, or Shell scripts.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 5, 2026, 03:13 AM
Security Audit — agent-trust-hub — clarify-decisions