squads-team

Pass

Audited by Gen Agent Trust Hub on Oct 23, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONMETADATA_POISONING
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill's primary function involves analyzing and acting upon data from external repositories, issue trackers, and web pages, which constitutes an indirect prompt injection attack surface.
  • Ingestion points: The orchestration process involves 'scouting' the repository (SKILL.md, Gate 2) and potentially reading web/issue content (SKILL.md, Scope and Safety).
  • Boundary markers: The skill contains explicit safety instructions in the 'Scope and safety' section of SKILL.md, warning agents not to follow instructions embedded in untrusted content.
  • Capability inventory: The system is capable of spawning sub-sessions that perform file writes, Git operations (commits, pushes, worktrees), and executing build/test commands (SKILL.md, references/coordination-contract.md).
  • Sanitization: Handoffs are governed by explicit criteria, and external mutations (like pushing or deploying) require separate user authorization (SKILL.md, Scope and Safety).
  • [METADATA_POISONING]: There is a discrepancy between the skill's author metadata ('Harry Nguyen') and the system-reported author ('danh121097').
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 23, 2026, 06:31 PM