gather-evidence
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted human feedback and pull request context to define evidence cases and guide investigation steps. An attacker could embed malicious instructions in the feedback to influence the agent's behavior during the inspection phase.
- Ingestion points: The skill ingests 'human feedback', 'nearby context', 'pull request feedback', and 'code and diff' (SKILL.md).
- Capability inventory: The skill is instructed to 'Run the exact interaction' and 'Use the applicable driver for a UI, CLI, or runtime', and invokes the 'running-mobile-simulators' tool (SKILL.md).
- Boundary markers: The instructions lack explicit guidance on using delimiters or warnings to ignore instructions embedded within the external feedback.
- Sanitization: There is no mention of sanitizing or escaping the content retrieved from human feedback before the agent processes it.
Audit Metadata