wayfinder

Pass

Audited by Gen Agent Trust Hub on Aug 5, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through its 'Research' ticket functionality. This feature utilizes subagents to ingest data from external, untrusted sources such as third-party documentation and APIs to resolve project decisions. This ingested content is then integrated into the project's 'Map' and 'Decisions so far', providing a vector for external data to influence the agent's logic or future planning steps.
  • Ingestion points: Data retrieved by /research subagents from external documentation, APIs, and local knowledge resources as described in the 'Ticket Types' section of SKILL.md.
  • Boundary markers: The instructions do not define specific delimiters or 'ignore' instructions for the data returned by research subagents before it is appended to the shared map.
  • Capability inventory: The skill can modify issues on a tracker, assign users, close tickets, and trigger further sub-skills or manual tasks based on the findings.
  • Sanitization: There are no explicit requirements for sanitization or validation of external research findings before they are utilized to update the project's canonical artifacts.
  • [COMMAND_EXECUTION]: The skill authorizes the agent to autonomously perform environment-altering manual work under the 'Task (AFK)' ticket type. These tasks include 'provisioning access' and 'moving data' to unblock project decisions, which may involve shell command execution or system-level modifications without immediate human oversight.
  • Evidence: The 'Ticket Types' section in SKILL.md explicitly allows 'Task' tickets to be driven 'by the agent alone where it can (AFK)' for tasks like provisioning and data movement.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 5, 2026, 08:49 AM
Security Audit — agent-trust-hub — wayfinder