architecture-review
Pass
Audited by Gen Agent Trust Hub on Jun 27, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill instructions and supporting files do not contain any malicious patterns, obfuscation, or unauthorized access attempts. All operations are local and consistent with the stated purpose of documenting architecture decisions.
- [COMMAND_EXECUTION]: The skill instructs the agent to create directories and write Markdown and HTML files to a local directory (defaulting to
./skill-outputs/architecture-review/). These file system operations are standard for a documentation tool and do not involve privileged or risky command execution. - [PROMPT_INJECTION]: The skill processes user-provided constraints and context to brief subagents. This surface is managed through structured briefings and isolated execution of personas, which naturally limits the risk of indirect prompt injection.
- Ingestion points: User input provided during the initial elicitation step in SKILL.md Step 0 (decision details, constraints, context).
- Boundary markers: The workflow uses a specific "Decision Brief" format in Step 1 to structure data before passing it to subagents, providing natural delimiters.
- Capability inventory: Capability is limited to local file system writes and directory creation; no network access or arbitrary code execution tools are utilized.
- Sanitization: The skill relies on standard agent handling of interpolated text within defined Markdown templates.
Audit Metadata