agent-team-orchestration-v3-public

Pass

Audited by Gen Agent Trust Hub on May 17, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: Includes a shell script (scripts/setup-team.sh) designed to automate the creation of agent workspaces. The script performs local directory creation and symlinking within the platform's standard configuration path (~/.openclaw).
  • [PROMPT_INJECTION]: The skill's orchestration architecture creates an indirect prompt injection surface where sub-agent outputs influence the main agent's decisions.
  • Ingestion points: The orchestrator reads agent-generated reports such as score-report.md, as detailed in references/workflow.md.
  • Boundary markers: Role-specific TOOLS.md templates in references/role-templates.md provide explicit constraints to prevent agents from reading their own instructions or entering tool-call loops.
  • Capability inventory: The main agent is granted sessions_spawn permissions to manage sub-agent lifecycles, as shown in references/architecture.md.
  • Sanitization: The workflow implements a structured parsing approach where the orchestrator extracts JSON from agent reports to determine success criteria, minimizing the risk of natural language instruction overrides (references/workflow.md).
Audit Metadata
Risk Level
SAFE
Analyzed
May 17, 2026, 10:32 PM
Security Audit — agent-trust-hub — agent-team-orchestration-v3-public