Self-Evolving Five-Layer Memory System

ReviewAudited by ClawScan on May 10, 2026.

Overview

The skill is a coherent memory framework, but it asks the agent to persist sensitive profile/config data and automatically evolve its own instruction files without clear user controls.

Review this carefully before installing. Use it only if you want a persistent local memory system, disable or manually gate automatic heartbeat/consolidation behavior, do not store secrets such as API keys, and require approval before it updates AGENTS.md, SOUL.md, or other instruction files.

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

Private profile details, contacts, conversation-derived preferences, or secret-related configuration could be stored and reused in future tasks.

Why it was flagged

The skill directs automatic extraction into persistent memory and explicitly lists sensitive identity/contact data and API-key-related environment facts as KG content, without clear redaction, retention, or user-approval controls.

Skill content
每次心跳执行:1. 对话记忆主动提取 → 写入 MemPalace KG ... 推荐 KG 实体分类:用户身份(称呼、OpenID、联系方式) ... 环境状态(组件、API Keys)
Recommendation

Only store explicitly approved memory, avoid storing API keys or secrets, add retention/redaction rules, and require user confirmation before writing sensitive entries.

What this means

Mistaken or manipulated memory could become a standing rule that changes how the agent behaves later.

Why it was flagged

Repeated error records can be promoted into persistent instruction files that govern the agent, so a bad classification or poisoned memory item could affect future sessions and decisions.

Skill content
进化机制:成长箱同类错误 3 次 → 晋升为规则 → 写入 AGENTS.md / SOUL.md
Recommendation

Require explicit human review and a visible diff before modifying AGENTS.md, SOUL.md, or other instruction-bearing files; keep changes reversible.

What this means

If a user later supplies or obtains the missing script, it could scan and mutate workspace memory files without the behavior being reviewed as part of this skill.

Why it was flagged

The artifacts reference and recommend automatic execution of a consolidation script, but the provided skill package is instruction-only and does not include that script for review.

Skill content
python3 scripts/mempalace_consolidation.py ... 建议在 HEARTBEAT.md 中配置为每次心跳自动执行。
Recommendation

Do not configure automatic execution until the script source is present, reviewed, pinned to a trusted source, and restricted to approved paths.