architect
Pass
Audited by Gen Agent Trust Hub on Sep 2, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted external data (existing subsystems and codebases) during its 'Ground' and 'Sketch' phases.
- Ingestion points: Phase A ('Ground the problem') explicitly instructs the agent to analyze relevant subsystems and existing structures using the
howskill to build a mental model. - Boundary markers: The skill does not define specific delimiters or instructions to ignore embedded prompts within the source material it analyzes.
- Capability inventory: The skill possesses the capability to execute other skills (
how,why,arena,interrogate) and perform file system writes in Phase D ('Implement against the sketch') where it replaces placeholders with functional code. - Sanitization: No explicit sanitization or validation of the ingested code/data is mentioned before it is processed by the agent or passed to the 'arena' runners.
Audit Metadata