nl-router
Pass
Audited by Gen Agent Trust Hub on Jun 18, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill acts as a natural language router, parsing untrusted user input to execute actions across various workflows, which creates an indirect prompt injection surface.
- Ingestion points: Natural language requests matched against trigger patterns defined in
SKILL.md(e.g., SDLC workflows, research investigations). - Boundary markers: None present; the patterns directly interpolate user-provided strings into route parameters without delimiters or 'ignore' instructions.
- Capability inventory: The router can trigger potentially sensitive workflows such as
security-assessment,incident-triage, andartifact-orchestration(which may involve file generation). - Sanitization: There is no evidence of validation or sanitization for parameters extracted from user requests (e.g.,
{task},{artifact},{phase}). - [SAFE]: The skill triggers related to instruction clarification (e.g., 're-read my instructions', 'go back and read what I wrote') are legitimate error-recovery mechanisms designed to improve agent accuracy rather than bypass safety filters.
- [SAFE]: No executable code, remote script downloads, sensitive file access, or network exfiltration patterns were detected in the skill configuration.
Audit Metadata