← 返回 Skills 市场
246
总下载
1
收藏
0
当前安装
2
版本数
在 OpenClaw 中安装
/install memory-attention-router
功能描述
Deterministic long-term memory routing for OpenClaw. Route, write, reflect on, and refresh reusable memory for multi-step agent work. Use when the task depen...
安全使用建议
This skill is internally consistent with its description, but review these practical points before installing: (1) It stores persistent memories in a local SQLite DB (default: <workspace>/.openclaw-memory-router.sqlite3); set MAR_DB_PATH to control the file location. Treat that DB as potentially sensitive data — back it up, restrict access, or encrypt if needed. (2) The agent will execute the included Python script (python3 scripts/memory_router.py); inspect the code if you do not trust the source. (3) There are no network calls or secret env vars requested, which reduces exfiltration risk, but be mindful that any persisted memory may include user-sensitive content. (4) If you want to restrict where data is written, create a dedicated directory and set MAR_DB_PATH accordingly. If you need more assurance, request a full review of the remainder of memory_router.py (the file is present in the package) or run the script in an isolated/test workspace first.
功能分析
Type: OpenClaw Skill
Name: memory-attention-router
Version: 1.1.0
The memory-attention-router skill is a legitimate utility designed to provide long-term memory management for OpenClaw agents using a local SQLite database. The Python implementation in `scripts/memory_router.py` follows security best practices by using parameterized SQL queries to prevent injection and contains no network-reaching code, obfuscation, or unauthorized file system access. The instructions in `SKILL.md` and the LLM prompt templates in `scripts/prompts/` are strictly functional, focusing on summarizing task outcomes and retrieving relevant context without any evidence of malicious prompt injection or data exfiltration logic.
能力评估
Purpose & Capability
Name/description match the included code and docs: the package implements deterministic routing, scoring, and CRUD for long-term memories using a local SQLite DB and produces compact 'packets' for agent steps. No unrelated credentials, binaries, or third-party services are requested.
Instruction Scope
SKILL.md instructs the agent to run the included Python script to route, add, reflect, refresh, and inspect memories. The runtime flow only reads/writes the local SQLite DB, uses deterministic routing rules, and references debug fields. It does not instruct network calls or reading unrelated system config. Note: it persists potentially sensitive memory content to disk (the SQLite DB) which is expected for this functionality; consider privacy implications.
Install Mechanism
No install spec or external downloads are used—the skill is delivered as local files (Python + SQL + docs). This is low-risk compared with remote installers. The agent will execute the provided Python script (so review code before allowing execution).
Credentials
The skill declares no required env vars or credentials, which is consistent with a local-only router. It does reference an optional MAR_DB_PATH environment variable (and a default DB location inferred from workspace layout) but does not require secrets; this is reasonable. (Minor point: MAR_DB_PATH is referenced in SKILL.md/code but not listed as a required env var—it's optional, not a credential.)
Persistence & Privilege
always:false (not force-enabled). The skill persists long-term memories in a SQLite DB in the workspace by default (or at MAR_DB_PATH if set). That persistence is expected for a memory router; it does not modify other skills' configs or request system-wide privileges. Ensure you are comfortable with creating/keeping a local DB containing user data.
如何使用
- 确保已安装 OpenClaw（本地或 Docker 部署）
- 在对话框中输入安装命令：
/install memory-attention-router - 安装完成后，直接呼叫该 Skill 的名称或使用
/memory-attention-router触发 - 根据 Skill 的参数说明提供必要输入，即可获得结构化输出
版本历史
v1.1.0
- Improved SKILL.md for clarity and conciseness, with updated descriptions and trigger cues.
- Added new trigger phrases like "replace my previous rule" and guidance on when to trigger for agent steps.
- Clarified that executor steps should preserve hard constraints alongside reusable procedures.
- Defined stricter caps for packet sizes to ensure compact, high-signal working-memory packets.
- Expanded routing rules, including use of support/contradicts edges and priority for durable, reusable memories.
- Refined write, reflect, and refresh flow instructions for better memory lifecycle management.
v1.0.0
- Initial release of the memory-attention-router skill for managing agent long-term memory in multi-step OpenClaw tasks.
- Routes, writes, reflects on, and refreshes memories based on user preferences, prior sessions, and workflow patterns.
- Triggers on explicit preference-memory phrases (e.g., "from now on", "always", "avoid").
- Builds compact, role-aware working-memory packets instead of dumping large memory lists or using plain document retrieval.
- Includes deterministic routing, replacement and retirement of stale memories, and clear role/task selection behaviors.
- Provides command-line workflows for routing, adding, reflecting, refreshing, and inspecting memories.
元数据
常见问题
memory-attention-router 是什么？
Deterministic long-term memory routing for OpenClaw. Route, write, reflect on, and refresh reusable memory for multi-step agent work. Use when the task depen... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 246 次。
如何安装 memory-attention-router？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install memory-attention-router」即可一键安装，无需额外配置。
memory-attention-router 是免费的吗？
是的，memory-attention-router 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
memory-attention-router 支持哪些平台？
memory-attention-router 跨平台运行，可在任意部署了 OpenClaw / Claude Code 的环境中使用（cross-platform）。
谁开发了 memory-attention-router？
由 Kai（@kaiqiangh）开发并维护，当前版本 v1.1.0。
推荐 Skills