Create SubAgent

PassAudited by ClawScan on May 10, 2026.

Overview

This is a coherent instruction-only helper for creating SubAgents, but users should deliberately scope and monitor any child agents it creates.

This skill appears safe to install as an instruction-only SubAgent helper. Before creating a SubAgent, choose run mode unless you need a persistent session, clearly limit the SubAgent’s task and file scope, monitor logs, and use the documented kill command if it behaves unexpectedly.

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.

What this means

A SubAgent could read or modify files, fetch web content, or run Python if the platform grants those tools.

Why it was flagged

The research and data SubAgent templates explicitly mention web access, file read/write, and Python execution tools. This is aligned with the stated purpose, but it means created agents can perform meaningful actions in the user’s environment.

Skill content
使用工具:web_search, web_fetch, read, write ... 工具:exec (Python), read, write
Recommendation

Give SubAgents narrow tasks, specify file/path limits when relevant, and review logs or outputs before relying on important changes.

NoteHigh Confidence
ASI10: Rogue Agents
What this means

A persistent SubAgent session may continue existing after creation until completed, archived, or killed.

Why it was flagged

The skill supports persistent SubAgent sessions and retained cleanup state, while also documenting user-facing termination commands. This is expected for a SubAgent manager but should be noticed.

Skill content
session - 持久会话 ... cleanup: keep ... /subagents kill <id|#|all>
Recommendation

Use the default run mode unless persistence is needed, and use list/log/kill commands to monitor or stop SubAgents.

What this means

Users may look for or run an external script that was not part of the reviewed artifacts.

Why it was flagged

The README documents a PowerShell script, but the supplied manifest contains only README.md, SKILL.md, and .clawhub.json, with no script included. This appears to be documentation/package inconsistency rather than automatic execution.

Skill content
.\create-subagent.ps1 -preset dev
Recommendation

Use the instruction-only workflow provided in SKILL.md, and do not run any separately obtained create-subagent.ps1 unless it is independently reviewed.