documenting-with-diagrams

Pass

Audited by Gen Agent Trust Hub on Jun 24, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill provides bash script recipes for rendering diagrams using the mmdc CLI tool. These operations are protected by a 'Pre-render safety' STOP gate in SKILL.md (Step 6), which requires the agent to verify tool availability and obtain user permission before execution or installation of dependencies.
  • [PROMPT_INJECTION]: The skill ingests untrusted source code or documentation to generate diagram specifications (Ingestion: SKILL.md Step 1). Although it lacks explicit boundary markers, this indirect prompt injection surface is mitigated by a mandatory 'Confirm inventory' STOP gate (SKILL.md Step 4), which requires human approval of the proposed diagram content before the agent exercises its Write capability or execution recipes.
  • [SAFE]: The instructions prioritize user oversight by ensuring that no files land on disk and no remote tools are invoked without preceding approval steps. The skill follows secure design patterns by enforcing user checkpoints for all potentially sensitive actions.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 24, 2026, 07:31 AM
Security Audit — agent-trust-hub — documenting-with-diagrams