paw-mkt-referral
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to audit live website URLs and local codebases (defined in references/shared-patterns.md), which serves as a significant ingestion point for untrusted data. This creates a potential surface where malicious instructions embedded in external content could influence the agent's behavior.
- Ingestion points: Live URL audits and existing codebase inspections described in the Starting Context Router.
- Boundary markers: The instructions do not specify any delimiters or warnings to ignore instructions embedded in the audited data.
- Capability inventory: The skill possesses file-write capabilities (saving deliverables to brand workspaces) and network access via agent-browser or WebFetch.
- Sanitization: No explicit content sanitization or validation logic is provided for external data inputs.
- [EXTERNAL_DOWNLOADS]: The skill references and provide instructions to install external tools such as 'agent-browser' from the 'vercel-labs' GitHub repository. While this is a trusted source, the skill's reliance on external script installation is an identified pattern.
- [COMMAND_EXECUTION]: The skill utilizes shell scripts for tool discovery and browser profile management (e.g., tool-discovery.sh, chrome-profiles.sh) to facilitate its research and automation capabilities within the local environment.
Audit Metadata