spec-driven-dev-v2
Pass
Audited by Gen Agent Trust Hub on May 20, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill is a project management framework and does not contain any malicious code or behavior. It uses local bash scripts (
scripts/task-lint.shandscripts/governance-check.sh) to enforce project invariants and validate task structures. These scripts rely on standard command-line utilities likejq,yq,awk, andgrepand do not perform network operations or access sensitive system paths. - [PROMPT_INJECTION]: The skill's architecture includes a review workflow that constitutes an 'Indirect Prompt Injection' surface, as a subagent (Reviewer) is tasked with processing untrusted data (git diffs and test outputs) generated by a builder agent. This is a functional requirement of the framework. The skill implements design-level mitigations by recommending 'reasoning-blind' reviewers and using markdown code blocks as boundary markers in the 'Context Pack' (defined in
templates/review/REVIEW.md) to encapsulate external data. - Ingestion points:
templates/review/REVIEW.md(Sections A.4 and A.6) ingest code diffs and command outputs. - Boundary markers: Markdown code blocks are used to delimit untrusted content.
- Capability inventory: The Reviewer subagent is restricted to evaluation and feedback; automated governance is handled by local scripts.
- Sanitization: The skill mandates that the Reviewer subagent be fresh and 'reasoning-blind' to prevent context contamination.
Audit Metadata