feature-brief
Pass
Audited by Gen Agent Trust Hub on Oct 1, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill provides structured instructions for a product management workflow, focusing on documentation and internal coordination within a development repository.
- [INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it is instructed to ingest untrusted data from the repository (such as 'usage data', 'affected code', and 'deployment configuration').
- Ingestion points: Analyzes repository files, deployment logs, and performance metrics via local file access.
- Boundary markers: Instructions mandate labeling all gathered information as 'observed', 'reported', or 'assumed' to distinguish sources.
- Capability inventory: Limited to writing Markdown files to a specific local directory (
.work/features/) and referring tasks to other internal agent skills; no arbitrary command execution or network exfiltration capabilities are present. - Sanitization: Relies on human-in-the-loop review ('the human decides', 'incorporate accepted feedback') and explicit evidence labeling to mitigate data poisoning risks.
Audit Metadata