multi-frontier-desktop
Pass
Audited by Gen Agent Trust Hub on Aug 24, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [SAFE]: The skill defines a set of security-focused operational guidelines for coordinating multiple AI agents in a desktop environment. It explicitly forbids the collection or persistence of credentials and enforces a 'least privilege' model for workspace modifications.- [PROMPT_INJECTION]: The skill establishes a multi-agent workflow that involves sharing outputs (proposals) and repository content between different AI models. This design creates a surface for indirect prompt injection where malicious data in a repository or one agent's proposal could influence the behavior of the other agent.
- Ingestion points: The workflow ingests user requests, repository data, and cross-agent proposals during the review phase (SKILL.md).
- Boundary markers: The instructions specify using 'bounded proposal artifacts' and 'bounded prompts,' although specific delimiter formats are not defined to ensure the agent reliably distinguishes instructions from data.
- Capability inventory: The system manages 'workspace_write' permissions and interacts with CLI tools such as 'codex' and 'claude'.
- Sanitization: The skill mandates IPC boundary verification and ensures that the renderer process cannot issue direct system commands.
Audit Metadata