adversarial-phoenix-liveview

Pass

Audited by Gen Agent Trust Hub on Jul 20, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: No malicious patterns, hardcoded credentials, or unauthorized network operations were found in the skill's instructions or supporting files.
  • [PROMPT_INJECTION]: The skill contains an indirect prompt injection surface as it ingests untrusted source code or diffs and interpolates them into a subagent's prompt for evaluation.
  • Ingestion points: User-provided code is interpolated into the {{TARGET_CONTENT_OR_PATHS}} field within references/reviewer-prompt.md.
  • Boundary markers: Absent. The reviewer prompt does not use hardened delimiters (like random XML tags) or explicit instructions to the subagent to ignore embedded commands within the target code.
  • Capability inventory: The subagent's primary capability is generating a pass/fail verdict report based on the provided code; it does not have access to restricted tools or network resources.
  • Sanitization: The skill does not perform sanitization or escaping of the target code before interpolation, which could allow maliciously crafted code comments to influence the reviewer subagent's final verdict.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 20, 2026, 06:56 AM
Security Audit — agent-trust-hub — adversarial-phoenix-liveview