workbench-agent-install-unifier

Pass

Audited by Gen Agent Trust Hub on May 1, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill requires the agent to read documentation and instructions from an untrusted external GitHub repository: https://github.com/millionco/agent-install.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it instructs the agent to ingest and follow instructions from external sources that are not from a verified or well-known vendor.
  • Ingestion points: The skill identifies https://github.com/millionco/agent-install and docs/agent-install-unifier-lane.md as required source material for its operation.
  • Boundary markers: No specific delimiters or "ignore previous instructions" warnings are defined for the processing of these external inputs.
  • Capability inventory: The skill description and rules indicate the capability to mutate local configuration files (e.g., AGENTS.md, MCP configs).
  • Sanitization: The rules section includes a prohibition on writing secrets and credentials, which provides a policy-level guardrail but lacks technical sanitization of the input data.
Audit Metadata
Risk Level
SAFE
Analyzed
May 1, 2026, 11:55 PM