architect

Pass

Audited by Gen Agent Trust Hub on Sep 8, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill defines a process where external data (source code, system documentation, and environment artifacts) is ingested as "grounding artifacts" in Phase A and passed to multiple AI "runners" in Phase B. This setup presents an indirect prompt injection surface, as malicious instructions hidden within a project's source code or documentation could attempt to influence the synthesized architecture or subsequent code generation.
  • Ingestion points: System grounding artifacts gathered in Phase A (via the 'how' and 'why' skills) and user-supplied architectural tasks.
  • Boundary markers: The references/runner-prompt.md template, which orchestrates the candidate models, does not currently use specific delimiters or explicit "ignore embedded instructions" directives for the grounding data it processes.
  • Capability inventory: The skill has the capability to modify the filesystem (Phase D: Implementation) and execute other complex agent skills (arena, harness, interrogate, how).
  • Sanitization: No explicit sanitization or filtering of the input grounding artifacts is performed prior to their inclusion in the design prompts.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 8, 2026, 06:20 AM
Security Audit — agent-trust-hub — architect