architect
Pass
Audited by Gen Agent Trust Hub on Nov 8, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes Python scripts and Git commands via subprocess.run to manage the architecture sealing process and workspace state. Inputs like the workspace slug are normalized to prevent path traversal, and script paths are validated to remain within the skill directory.
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied architecture drafts. 1. Ingestion points: {input.draft} file in SKILL.md. 2. Boundary markers: present in delegation-protocol.md advising the agent to ignore and report embedded instructions. 3. Capability inventory: subprocess.run in architecture_handoff.py and architecture_workspace.py, and atomic file writes in seal_assessment.py. 4. Sanitization: present in architecture_contract.py via strict structure, line-range, and citation validation.
- [DYNAMIC_EXECUTION]: _skill_protocol_runtime.py uses importlib.import_module for dynamic loading of the diagnostic contract helper. The module names are hardcoded internal strings, posing no risk of arbitrary code execution from user input.
- [SAFE]: The skill modifies .git/info/exclude to ensure generated architecture artifacts are ignored by Git. This is a standard project management practice consistent with the skill's stated purpose.
Audit Metadata