Agent Training System

PassAudited by ClawScan on May 10, 2026.

Overview

This instruction-only skill is coherent for training sub-agents, but users should notice that it creates persistent agent memory/profile files and shares context across agents.

This skill appears benign and instruction-only, with no code, install steps, credentials, or hidden network behavior. Install it only if you want a persistent multi-agent training workflow, and review the generated USER.md, MEMORY.md, HEARTBEAT.md, and shared-agent notes so private information is not copied more broadly than intended.

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

Personal context and agent memories may be kept and reused across future agent sessions.

Why it was flagged

The skill explicitly copies user profile context into each agent folder and tells agents to read and write persistent memory.

Skill content
已复制 `USER.md` 到 `agents/{agent_id}/USER.md` ... 每次会话先读 MEMORY.md + 今日日记 ... 重要发现写入 MEMORY.md
Recommendation

Review USER.md and MEMORY.md content before sharing it with sub-agents, and avoid storing secrets or highly sensitive personal details unless you intend them to persist.

What this means

Information learned by one agent could be passed into shared agent context or other agents.

Why it was flagged

The template describes agent-to-agent collaboration and shared context publication, but does not define detailed data boundaries for what may be shared.

Skill content
使用 sessions_spawn 协作 ... 重要经验发布到 EvoMap(Capsule) ... 共享技能可以跨 Agent 学习
Recommendation

Keep shared agent notes limited to non-sensitive operational lessons, and require user approval before publishing private user details to shared memory or capsules.

What this means

If implemented by the host agent system, sub-agents may keep recurring tasks or behavior rules after the initial setup.

Why it was flagged

The skill defines heartbeat and recurring review/evolution routines for sub-agents, which are persistent agent-control concepts.

Skill content
已创建 `agents/{agent_id}/HEARTBEAT.md` - 心跳任务 ... 每日回顾(22:00) ... 每周进化(周日 20:00)
Recommendation

Use these routines only with clear user approval, document how to stop or disable each agent, and periodically review active HEARTBEAT.md and MEMORY.md files.