triaging-visual-review-runs

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.80). The skill instructs the agent to fetch and embed pre-signed S3 download URLs verbatim (e.g., in curl commands) so the LLM would need to handle and potentially output time-limited secret URLs that grant access to artifacts, creating an exfiltration risk.

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.70). The skill's "Visual inspection" workflow instructs the agent to fetch and "Read" baseline/current PNG files from the run's artifact download URLs (current_artifact.download_url / baseline_artifact.download_url — presigned S3 URLs) and to use what it sees to form triage verdicts, so it clearly ingests untrusted, user-generated third-party content that can influence decisions.