task
Pass
Audited by Gen Agent Trust Hub on Jun 18, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill is designed for local project management. It reads a local file (
spec.md) and writes to a local file (tasks.md) without requesting network access or executing external code. - [PROMPT_INJECTION]: The skill processes external input from
spec.md, which creates an interface for indirect prompt injection. However, the design includes a human-in-the-loop 'Progressive Disclosure' mechanism requiring user approval for each phase, which significantly mitigates the risk of automated instruction overrides. - Ingestion points:
spec.md(read operation) - Boundary markers: None explicitly defined to isolate input data
- Capability inventory: Local file writing (
tasks.md) - Sanitization: No explicit content validation identified
Audit Metadata