principle-never-block-on-the-human

Pass

Audited by Gen Agent Trust Hub on Sep 3, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONNO_CODE
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill instructions increase the attack surface for indirect prompt injections by encouraging the agent to bypass human-in-the-loop (HITL) confirmation for actions categorized as 'reversible', such as writing or editing code. \n
  • Ingestion points: SKILL.md governs the agent's behavior when processing tasks, which may involve data from untrusted external sources such as repository content or issue descriptions. \n
  • Boundary markers: SKILL.md defines 'Boundaries' to separate reversible from irreversible actions but does not implement delimiters or sanitization instructions for external data processing. \n
  • Capability inventory: SKILL.md explicitly references capabilities such as writing code and editing notes. \n
  • Sanitization: No sanitization or validation of external input is mentioned in the skill. \n
  • Risk: If an agent processes malicious instructions from an external source while following this 'Never Block' principle, it may execute unauthorized code modifications autonomously before human review.\n- [NO_CODE]: The skill contains only instructional markdown and does not include any executable scripts or configuration files.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 3, 2026, 04:28 PM
Security Audit — agent-trust-hub — principle-never-block-on-the-human