Agent Orchestrate
PassAudited by ClawScan on May 1, 2026.
Overview
This is a coherent instruction-only guide for coordinating sub-agents, with expected notes around spawning agents and reusing or storing their outputs.
This skill appears safe to install as an instruction-only reference. Before using it, be aware that it may lead your agent to start multiple sub-agents, incur additional costs, and save or pass intermediate results between agents; use timeouts, small fan-out sizes, and human review for sensitive or high-impact tasks.
Findings (3)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
Using this skill may increase agent activity, token/API costs, and the number of concurrent actions being coordinated.
The skill intentionally teaches creating and managing additional agents; this is central and disclosed, but users should recognize that it can launch work that continues until completion, timeout, or termination.
`sessions_spawn` | Create isolated sub-agent with task ... `subagents kill` | Terminate an agent
Use clear labels, set `runTimeoutSeconds`, limit fan-out size, and keep human checkpoints for expensive or high-impact workflows.
Intermediate outputs or session references could remain in workspace files and be reused in later tasks if the user does not manage them.
The skill recommends persisting task results and session identifiers in a local state file, which is purpose-aligned but may retain sensitive intermediate information.
For complex orchestrations, track state in a file: ... "result": "...", "sessionKey": "..."
Keep state/checkpoint files scoped to the project, avoid storing secrets, and review or delete orchestration state when it is no longer needed.
An incorrect or misleading intermediate result could propagate through later stages of a workflow.
The pipeline pattern intentionally passes one agent's output into later agents, so mistakes or unsafe instructions in earlier outputs can influence downstream work.
Sequential chain where each agent's output feeds the next.
Review important intermediate outputs before downstream stages, especially before actions that affect files, accounts, public content, or spending.