discuss-task
Pass
Audited by Gen Agent Trust Hub on Oct 27, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted external data, such as codebases, documentation, and error logs, to identify material uncertainty and build decision trees. This creates a surface where embedded malicious instructions in the analyzed files could attempt to influence the agent's reasoning or the questions it asks.
- Ingestion points: SKILL.md (Step 2) and INTERVIEW.md explicitly instruct the agent to inspect discoverable context including files, docs, and data.
- Boundary markers: The instructions do not explicitly define boundary markers for external data, though they emphasize separating discoverable facts from user decisions.
- Capability inventory: The skill can spawn sub-agents (deep-thinker), inspect local file systems, and route to other internal skills (to-prd, five-whys).
- Sanitization: No specific sanitization or filtering of external content is mentioned.
- Mitigation: The skill includes a robust mitigation by requiring the agent to summarize the shared understanding and obtain explicit user confirmation before acting on any resulting plan or routing to the next skill.
Audit Metadata