Naruto Multi-Agent CN

ReviewAudited by ClawScan on May 10, 2026.

Overview

It does not show malware or credential theft, but it needs review because it routes all real work into persistent sub-agent sessions that can retain and mix task context.

Install only if you want this persistent multi-agent workflow. Do not use it for sensitive or secret-heavy work unless you can control what context is forwarded and clear or isolate the sub-agent sessions afterward.

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

When active, the skill may stop the main agent from doing tasks directly and route them through sub-agents instead.

Why it was flagged

This explicitly changes the main agent’s role and forces a specific delegation tool pattern for real work.

Skill content
你是纯调度员。你不能使用 exec、文件读写、搜索等任何执行工具。所有实际工作必须通过 `sessions_spawn` 委派给忍者。
Recommendation

Use this only when you intentionally want a multi-agent dispatcher; do not invoke it for tasks where you expect direct single-agent execution.

What this means

Sensitive context or prior instructions from one task could persist and influence later unrelated tasks handled by the same sub-agent.

Why it was flagged

The skill intentionally reuses fixed sessions so sub-agents remember prior work, but it does not describe reset, retention, or isolation controls.

Skill content
`sessionKey` 的作用是让同一个忍者复用同一个 session,保持记忆连续!
Recommendation

Avoid using this with secrets or unrelated projects unless you can clear or isolate the spawned sessions; add explicit reset/retention guidance.

What this means

Private user details may be forwarded into one or more persistent sub-agent sessions and retained there beyond the immediate request.

Why it was flagged

The coordinator must copy needed conversation/task context into spawned agent messages, but the artifacts do not define data-minimization or sensitive-context boundaries.

Skill content
task — 自包含的任务卷轴(忍者看不到你和委托人的对话,必须写清楚)
Recommendation

Forward only the minimum necessary context, ask before spawning on sensitive tasks, and avoid including passwords, tokens, private files, or unrelated conversation history.