risk-calibrated-pr-review
Pass
Audited by Gen Agent Trust Hub on Oct 1, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONPROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted PR data such as code diffs, implementation details, and author comments. This creates a potential surface where embedded instructions could attempt to influence the agent's verdict. Ingestion points are defined in the
INPUTsection ofSKILL.md. While the skill does not specify boundary markers or sanitization, it explicitly mandates a read-only boundary for the agent, which limits the potential impact of such an injection. - [PROMPT_INJECTION]: The directive 'CONVERGENCE LAW — OVERRIDES EVERY LATER INSTRUCTION' in
SKILL.mdemploys override language to ensure the agent follows the specified review logic over subsequent prompts. This is a common instructional hardening technique for complex multi-turn tasks but utilizes patterns typical of prompt injection.
Audit Metadata