team-qa
Pass
Audited by Gen Agent Trust Hub on Sep 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from external story files and project status reports (e.g., production/sprints/) and passes it to subagents (qa-lead, qa-tester). This interpolation lacks explicit boundary markers or sanitization, allowing malicious instructions within the data to potentially influence agent behavior.
- Ingestion points: Story files and status documents are read during Phase 1 and Phase 4.
- Boundary markers: No specific delimiters or "ignore instructions" warnings are defined for the untrusted content being passed to subagents.
- Capability inventory: The skill has access to the Write tool for file modifications and the Task tool for spawning additional agents, which could be abused if an injection is successful.
- Sanitization: The workflow relies on manual user approval gates via AskUserQuestion rather than automated sanitization of the input data.
Audit Metadata