agent-squad

Pass

Audited by Gen Agent Trust Hub on Sep 6, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The orchestration architecture implements a complex data pipeline where information is passed between specialized sub-agents. This structure creates a multi-step injection surface if any agent in the chain processes untrusted external data.\n
  • Ingestion points: The Rex agent (rex/SKILL.md) is responsible for extracting user intent and platform context, while the Main Agent (SKILL.md) coordinates and relays reports across the entire squad.\n
  • Boundary markers: The skill employs structured communication protocols, using 'Briefing Packets' and 'Project State' objects to maintain logical separation between agent contexts and tasks.\n
  • Capability inventory: The squad includes agents with high-privilege capabilities, such as Mason (mason/SKILL.md) for code generation, Dep (dep/SKILL.md) for CI/CD and Infrastructure-as-Code generation, and Quinn (quinn/SKILL.md) for test execution.\n
  • Sanitization: The 'Main Agent' implements a 'Context Window Discipline' that involves compressing and summarizing reports before they are passed to the next agent. This summarization serves as a structural filter, although there is no explicit instruction-level sanitization mentioned for the relayed content.\n- [COMMAND_EXECUTION]: The skill suite includes instructions for performing system-level operations during project setup and verification.\n
  • Evidence: The Quinn agent (quinn/SKILL.md) is tasked with 'executing comprehensive test suites', including integration tests that interact with real databases and external APIs. The Mason agent (mason/SKILL.md) is responsible for project initialization using package managers like npm or pip and setting up local development environments.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 6, 2026, 07:15 AM
Security Audit — agent-trust-hub — agent-squad