phase-6-ui-integration
Pass
Audited by Gen Agent Trust Hub on Aug 31, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill prescribes a 'Zero Script QA' process where the agent validates implementation by analyzing runtime logs that include responses from external APIs. This creates an attack surface where malicious data from a compromised backend could attempt to influence the agent's behavior.
- Ingestion points: The agent is instructed to review runtime logs ('[RESPONSE]') and the 'ui-qa.md' deliverable as part of the check phase.
- Boundary markers: The guidance does not specify the use of delimiters or 'ignore' instructions when processing external data in logs.
- Capability inventory: The skill allows access to powerful tools including Bash, Write, and Edit, which the agent could be directed to use if an injection occurs.
- Sanitization: No strategies for sanitizing or escaping external API content are provided for the QA review phase.
Audit Metadata