benchmark-review-configs
Pass
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- Local Command Execution: The skill instructs the agent to execute a bundled Node.js script located in the plugin directory. This pattern allows the skill to perform complex analysis while maintaining local scope, though it requires the execution environment to have Node.js available.
- Untrusted Data Ingestion: The skill processes external JSON and JSONL files provided by the user. This creates a surface for indirect prompt injection where data could contain instructions meant to deceive the agent.
- Ingestion points: Data enters the context via local file paths for review records defined in SKILL.md.
- Boundary markers: The instructions contain explicit directives to treat all input strings as inert data and ignore embedded instructions.
- Capability inventory: The skill uses the node runtime to execute a local script (reviewops.mjs) as specified in SKILL.md.
- Sanitization: The skill employs behavioral instructions to ensure the agent ignores non-data content and avoids following links or paths found within records.
Audit Metadata