closed-loop-delivery

Pass

Audited by Gen Agent Trust Hub on Apr 16, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection due to its processing of external feedback during the development loop.
  • Ingestion points: Untrusted data enters the agent context through Pull Request comments and reviews as specified in step 4 of the workflow in SKILL.md.
  • Boundary markers: The skill does not implement delimiters or instructions to ignore embedded commands within the fetched PR content.
  • Capability inventory: The skill executes arbitrary commands for local verification and development deployment as described in steps 3 and 5 of the workflow in SKILL.md.
  • Sanitization: No sanitization or validation of the PR comment content is performed before it is used to guide subsequent implementation and verification steps.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 16, 2026, 02:19 AM