cy-create-techspec
Pass
Audited by Gen Agent Trust Hub on Apr 15, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill's operations are confined to the local project environment, specifically within the
.compozy/directory. It uses a structured workflow (Gather context -> Ask questions -> Create ADRs -> Draft -> Review -> Save) that keeps the user in the loop for all significant decisions. - [SAFE]: Codebase exploration is performed using standard agent tool calls to understand architectural patterns, which is a routine and expected behavior for a technical design assistant.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from local PRD files and the existing codebase. While this is a theoretical attack surface for indirect prompt injection, the risk is mitigated by the mandatory interactive review phase where the user must manually approve the final draft before any files are written.
- Ingestion points:
.compozy/tasks/<name>/_prd.mdand project codebase files via exploration tools. - Boundary markers: Absent for untrusted file content.
- Capability inventory: File system read/write operations and directory creation restricted to the local project; codebase exploration via agent tools.
- Sanitization: None explicitly defined for PRD or codebase content.
Audit Metadata