team
Pass
Audited by Gen Agent Trust Hub on Nov 8, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill makes extensive use of shell commands to manage the orchestration environment. This includes using
tmuxfor process multiplexing,gitfor worktree isolation and branch management, andnodeto execute local maintenance scripts likecleanup-orphans.mjs. - [EXTERNAL_DOWNLOADS]: The documentation instructs users to install external CLI tools including
@openai/codex,@google/gemini-cli, and theantigravitytool from Google. It also referencespsmuxfrom a public GitHub repository. These tools are used as autonomous executors within the team structure. - [INDIRECT_PROMPT_INJECTION]: The orchestrator (lead agent) processes outputs from multiple sub-workers and external CLI tools to update the task list and generate context for subsequent pipeline stages. This creates a surface for indirect instructions to influence the lead agent's behavior.
- Ingestion points: The lead agent reads worker output files (e.g., in
.omc/research/), handoff documents in.omc/handoffs/, and direct messages from sub-agents. - Boundary markers: The skill uses structured Markdown templates for handoffs and JSON for inter-agent communication to delimit data.
- Capability inventory: The lead agent has full access to file read/write, shell command execution (via Bash and tmux), and the ability to spawn further sub-agents.
- Sanitization: The skill mentions the use of
sanitizeName()to prevent injection when creating Git branches based on worker or team names. - [DYNAMIC_EXECUTION]: The skill dynamically generates prompts (preambles) for sub-agents based on their assigned roles and tasks. It also manages the lifecycle of external workers by spawning them into dedicated tmux panes with specific working directories.
Audit Metadata