workos-authkit-base
Pass
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [Indirect Prompt Injection] (LOW): The skill defines an external README as the 'source of truth' for implementation, exposing the agent to instructions from untrusted external data.
- Ingestion points: External README fetched via WebFetch (SKILL.md).
- Boundary markers: Absent; the agent is instructed to follow README patterns explicitly.
- Capability inventory: Subprocess execution (npm install, build) and file system modification.
- Sanitization: Absent; no validation is performed on the fetched instructions.
- [External Downloads] (LOW): The skill dynamically fetches documentation from npmjs.com or GitHub to derive commands, which is a potential risk if the target package name is manipulated or the source is compromised.
- [Command Execution] (LOW): The skill guides the agent to perform package installations and build verifications based on logic derived from external, non-static sources.
Audit Metadata