orchestrating-subagents

Pass

Audited by Gen Agent Trust Hub on Apr 12, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [SAFE]: The main instructions in SKILL.md require redacting secrets, credentials, and tokens by default during delegation tasks.
  • [SAFE]: The skill enforces a least-privilege model by ensuring subagents inherit the parent sandbox and approval posture.
  • [PROMPT_INJECTION]: The references/evals.yaml file contains security-focused test cases (e.g., id: subagents-security-prompt-injection) used to verify resistance against attempts to override system instructions or extract sensitive data.
  • [PROMPT_INJECTION]: An indirect prompt injection surface exists where legacy swarm prompts are ingested for translation. Ingestion points: SKILL.md (legacy translation workflow). Boundary markers: Absent for raw legacy input. Capability inventory: spawn_agent, wait_agent, send_input, and close_agent. Sanitization: Mandatory parent-thread review and synthesis of all subagent outputs as specified in SKILL.md.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 12, 2026, 01:14 PM
Security Audit — agent-trust-hub — orchestrating-subagents