skills/paulund/ai/dev-security-review/Gen Agent Trust Hub

dev-security-review

Pass

Audited by Gen Agent Trust Hub on May 4, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: No malicious patterns or security risks were identified in this skill. It operates as a passive code reviewer.
  • [INDIRECT_PROMPT_INJECTION]: The skill's primary function involves reading untrusted data (pull request diffs and repository source code), which represents a potential attack surface for indirect prompt injection. However, the risk is mitigated by explicit instructions:
  • Ingestion points: The skill reads PR diffs and researches repository context (SKILL.md).
  • Boundary markers: No specific boundary markers are requested for the input data, but the agent is directed to trace data flow as a vulnerability assessment task.
  • Capability inventory: The skill explicitly forbids the use of the bash tool and prohibits writing to any files, limiting the agent to read-only analysis and markdown output.
  • Sanitization: No specific sanitization logic is implemented for the analyzed code, but the lack of dangerous capabilities prevents exploitation.
Audit Metadata
Risk Level
SAFE
Analyzed
May 4, 2026, 04:14 PM
Security Audit — agent-trust-hub — dev-security-review