qa-loop
Pass
Audited by Gen Agent Trust Hub on Sep 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes potentially malicious data from logs, network payloads, and browser output, presenting a risk of indirect prompt injection. \n
- Ingestion points: SKILL.md (Phase 3) outlines the ingestion of application logs and network evidence. \n
- Capability inventory: The agent can start background processes and edit source code to apply fixes (SKILL.md, Phase 2 & 5). \n
- Boundary markers: The skill explicitly instructs the agent to treat these inputs as "untrusted evidence" rather than instructions. \n
- Sanitization: Instructions are provided to redact sensitive information like secrets and tokens before recording evidence.\n- [COMMAND_EXECUTION]: The skill manages local application lifecycles by initiating managed background processes based on the project configuration (SKILL.md, Phase 2).
Audit Metadata