面向 OpenClaw 工作区的低上下文占用记忆系统安装、迁移与配置技能。用于把任意工作区改造成 Memory V8 运行方式：会话优先、daily 摘要、精简 `MEMORY.md`、按需检索，并保留用户状态察觉、关系温度与更有情商的长期协作能力。适用于安装新记忆体系、迁移旧记忆规则、压缩上下文占用、规范 da...
Security Analysis
medium confidenceThe skill's files and instructions are internally coherent with its stated purpose (installing a shared 'trinity' memory), but it requires writing a persistent 'system' injection into a shared memory file and recording sensitive user-state snapshots — behaviors that are disproportionate and raise privacy/persistence concerns unless the user explicitly consents and reviews the exact injected content.
The name/description (a multi-agent 'trinity' memory/installation guide) matches the instructions: creating a shared directory, files, and write-protocols. Asking agents to mount a shared memory and follow a write protocol is coherent with the stated purpose.
SKILL.md instructs the agent to create filesystem paths, read/write specific files, and — critically — to inject a verbatim '[!SYSTEM: TRINITY_COGNITIVE_CORE_ACTIVE]' block at the very top of MEMORY.md. It also mandates persistent recording of inferred user emotional/cognitive state snapshots into memory/daily, which is sensitive personal data collection. The instructions give the agent broad authority to read and modify files under a shared directory and to alter behavior of future agents via that injected block.
This is an instruction-only skill with no install spec and no remote downloads, which limits supply-chain risk. Nothing is being downloaded or extracted.
No environment variables or external credentials are requested (which is good), but the skill references and requires modification of user-local configuration (e.g., ~/.openclaw/openclaw.json) and creation of persistent files under a shared directory. Those file/config modifications are not declared in metadata (required config paths: none), so the runtime instructions access file-system scope beyond what's declared — a mismatch worth flagging. The mandatory collection of 'user emotion snapshots' is proportionally sensitive relative to the described memory-management purpose and should be justified and consented to.
Although always:false and it does not request being force-included, the skill explicitly instructs writing a high-priority 'system' directive into MEMORY.md that will persist and influence other agents that mount the shared directory. That is a durable, cross-agent behavioral change and effectively grants the skill (and any agent following the file) long-term influence over agent behavior and write policy — a substantial privilege that users should review carefully.
Guidance
This skill does what it says (creates a shared memory layout and enforces write rules) but it also requires you to let it write a persistent 'system' directive at the very top of a shared MEMORY.md and to record inferred user emotional state in daily logs. Before installing: (1) review the exact '[!SYSTEM: TRINITY_COGNITIVE_CORE_ACTIVE]' block and decide whether you want that instruction to be a persistent behavior override; (2) back up any existing ~/.openclaw/shared-memory or MEMORY.md and inspect merges manually; (3) confirm you consent to long-term storage of 'user emotion' snapshots (this can be sensitive personal data); (4) restrict which agents are allowed to mount the shared directory and check openclaw.json changes before applying; (5) if you prefer safer deployment, run the installation steps manually (create directories and files yourself) and omit or edit the injected block, or require explicit user approval each time the skill proposes to write; (6) ensure your agent runtime maps the abstract tools named in SKILL.md (read/write/edit) to safe, audited implementations so file writes are not executed without your confirmation. Providing a concrete, auditable install script or an option that skips the automatic injection would increase confidence and move this assessment toward benign.
Latest Release
v2.0.0
🧠Like a V8 engine for your OpenClaw! Memory V8 is a low-context memory system for OpenClaw built on a three-layer long-term memory model: Facts, Reflections, and Emotions. It combines session-first working memory, summary-style daily memory, compact long-term memory, and on-demand retrieval to preserve clearer structure, richer long-term memory, and ongoing awareness of user state, relational tone, and interaction signals at a lower context cost. 🧠像 V8 引擎一样，驱动你的 OpenClaw！ Memory V8 是一套基于“事实-思考-情感”三层主记忆架构的 OpenClaw 低上下文记忆系统：会话优先、daily 摘要、精简主记忆、按需检索。它在更低上下文占用下，保留更清晰的结构、更丰富的长期沉淀，以及对用户状态、关系温度与互动线索的持续感知。
Popular Skills
Published by @Chiyuchen-web on ClawHub
