← 返回 Skills 市场
157
总下载
0
收藏
0
当前安装
14
版本数
在 OpenClaw 中安装
/install smart-memory-plus
功能描述
Complete memory system for OpenClaw agents. Combines enhanced memory management (WAL protocol, type classification, temporal decay, session cache) with conte...
安全使用建议
This package appears coherent and implements a local-only memory system. Before installing, consider: 1) Review and test scripts with the workspace pointed to a disposable directory (set OPENCLAW_WORKSPACE to a temp folder) to confirm behavior and outputs. 2) Keep a backup of any existing ~/.openclaw/workspace because the skill warns it conflicts with other memory skills (it will move/archive files). 3) The SKILL.md enforces that the agent must use the provided scripts for writes — this is a policy the skill expects, but the platform or agent could bypass it and write files directly, so only grant file-write capability if you trust the agent or run it sandboxed. 4) Use dry-run options (many scripts have --dry-run or list/status commands) to inspect what would be changed before allowing destructive operations. 5) Be cautious with the graph tool's raw-query mode: treat it as read-only tooling for inspection and avoid running untrusted SQL fragments against your workspace. If you want higher assurance, run the scripts manually in an isolated environment first.
功能分析
Type: OpenClaw Skill
Name: smart-memory-plus
Version: 1.4.0
The 'smart-memory-plus' skill bundle is a comprehensive and well-architected memory management system for OpenClaw agents. It features multi-layered security controls, including explicit path-traversal protections (assert_safe_path in compact_session.py), automated redaction of sensitive data like API keys and passwords via regex (session_state.py, extract_memories.sh), and restricted read-only SQL queries for its knowledge graph (memory_graph.py). The code is strictly local, utilizes only standard libraries, and contains no evidence of data exfiltration, unauthorized network calls, or malicious prompt injection.
能力评估
Purpose & Capability
Name/description (memory management, WAL, compaction, decay, search/graph) match the included scripts and references. No unrelated binaries or cloud credentials are requested; optional env vars only control the local workspace/session id. Files operate on MEMORY.md and workspace/memory/ which is appropriate for the stated purpose.
Instruction Scope
SKILL.md and the scripts limit all writes to the declared workspace and require the agent to use the provided scripts (WAL, compaction, extract, decay, graph) rather than doing direct file writes. The enforcement is procedural: scripts perform path checks and sanitization, but the agent platform could still bypass these scripts by writing files directly if allowed. Also note the memory_graph tool exposes a raw query mode (user-supplied WHERE clause / SQL fragment) which is consistent with a graph tool but should be used carefully as it can read memory contents.
Install Mechanism
No install spec — the skill is instruction-only with included scripts. No downloads or installs from external URLs; all code is present in the package. This is low risk from an install-mechanism perspective.
Credentials
No required credentials or sensitive environment variables are declared. The only env vars are optional workspace/session configuration (OPENCLAW_WORKSPACE, OPENCLAW_SESSION_ID), which are proportionate to the task. Scripts explicitly redact or reject common API key and private-key patterns.
Persistence & Privilege
The skill does not request always: true or other elevated platform privileges. It writes only to files under the declared workspace and /tmp session cache as documented. Scripts create/modify files in the workspace — expected for a memory system — and include logic to avoid overwrites and limit compacts to the last 30.
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 smart-memory-plus - 安装完成后，直接呼叫该 Skill 的名称或使用
/smart-memory-plus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
v1.4.0
feat: add heartbeat memory hygiene checks, session state isolation via --session-key, state files moved to memory/session-states/
v1.3.0
feat: session isolation via --session-key flag. Session state files now stored in memory/session-states/ directory instead of workspace root. Each session (Discord channel, Telegram DM) gets independent SESSION-STATE-<id>.md file.
v1.2.2
smart-memory-plus 1.2.2
- Updated scripts/memory_graph.py (details not specified)
- No changes to usage, requirements, or security documented
- All workflows, boundaries, and script interfaces remain as previously documented
v1.2.1
Smart-memory-plus v1.2.1 introduces incremental search index updates.
- Adds an incremental update command to memory_search_bm25.py for faster search index refreshes.
- SKILL.md updated: search quick reference now includes both full rebuild and incremental update commands.
- No changes to core workflows or security model.
v1.2.0
**Added experimental knowledge graph support for enhanced memory relations.**
- Introduced a new memory graph layer (`memory/.index/graph.db`) for entity-relation extraction and queries.
- Added `scripts/memory_graph.py` for building the knowledge graph, exploring related entities, querying, and graph stats.
- Updated reference and workflows to include graph-based actions (build, related, stats, raw queries).
- No external dependencies; all graph operations remain fully local and script-driven.
v1.1.1
smart-memory-plus v1.1.1
- Added new "Find related memories" capability with `classify_memory.py --related` command for relation-based memory search.
- Updated SKILL.md: expanded "Search & Relations" quick reference to document the new relation lookup feature.
- No external dependencies added; backward compatibility maintained.
v1.1.0
- Adds BM25-based semantic search: new script memory_search_bm25.py for advanced memory search.
- Updates documentation to include memory_search_bm25.py usage and workflows for building and querying the search index.
- No external dependencies introduced; all features remain fully local and privacy-safe.
v1.0.6
smart-memory-plus 1.0.6
- Adds a conflict warning: this skill now explicitly replaces both `smart-memory` and `context-compactor`. Do not install together—file conflicts will occur.
- Updates session start workflow: reading and using recent compacts for context is now a manual step, not automatic.
- Clarifies in the documentation that the skill does not auto-inject context from compacts; agents must process compact content themselves.
- Minor documentation clarifications throughout SKILL.md for safer usage and improved workflow description.
v1.0.5
**smart-memory-plus 1.0.5 Changelog**
- Improved context extraction accuracy in `compact_session.py` (better regex and edge case handling).
- Enhanced `extract_memories.sh` to skip empty or malformed lines, reducing noise in extracted memories.
- Refined temporal decay logic in `memory_decay.py` for smoother promotion and cleanup of long-term facts.
v1.0.4
**1.0.4 — Enforces use of memory management scripts for all writes**
- Requires agents to always use provided scripts for memory file writes; direct writes are forbidden and explicitly documented.
- Clarifies script responsibilities and forbidden actions in a new table for each memory location.
- Adds a critical warning: never use generic file-write tools directly on memory files—scripts must be used for all updates, enforcing sanitization, deduplication, and append-only operations.
- Rephrased and expanded documentation for security boundaries, reducing ambiguity. No changes to underlying code or logic; documentation only.
v1.0.3
**Summary:** Removes all LLM/cloud dependencies for memory extraction and compaction; now fully local and keyword/regex-based.
- Eliminated all LLM-based extraction/compaction modes — all scripts now operate fully locally with zero network or cloud requirements.
- Updated privacy boundaries: Compacts and scripts guarantee no external calls, no LLM usage, and strict data locality.
- Simplified extraction and compaction workflow references: Only keyword and regex-based methods remain (no more local LLM opt-in).
- Updated documentation, quick references, and agent behavior sections to reflect "local-only" operation.
- Security section and scripts table now explicitly highlight zero LLM and network usage.
v1.0.2
**Privacy and data security rules clarified; LLM usage restricted to local endpoints only.**
- Added explicit restriction: cloud LLM providers are disallowed for extraction and compaction; only local LLM (e.g. Ollama, LM Studio) is permitted.
- Clarified that all scripts and default keyword extraction are fully local and make zero network calls.
- Updated documentation in extraction and compaction sections to highlight privacy guarantees and proper agent configuration for local LLM.
- Strengthened warnings about cloud LLM usage potentially violating local-only data boundaries.
- No changes to workflows or APIs; documentation and policy updates only.
v1.0.1
**Minor documentation update clarifying LLM data boundaries.**
- Added explicit warning about cloud LLM usage in extraction and compaction modes: conversation data may be sent to the agent's configured external LLM provider.
- Clarified that keyword modes are fully local and involve zero network calls or token cost.
- Made it clear that the skill itself never performs network calls, but agent configuration may result in external data transmission when using LLM prompts.
- Minor documentation edits to explain `memory_search` is not provided by this skill.
v1.0.0
Smart Memory Plus 1.0.0 – initial release
- Provides unified memory management and context compression for OpenClaw agents with zero external dependencies.
- Implements multiple memory layers: hot session state, daily notes, curated long-term memories, compacts, archive, and session cache.
- Enforces strict write boundaries and automated sensitive data filtering for API keys, passwords, and private keys.
- Introduces a Write-Ahead Log (WAL) protocol to prevent context loss; all memory updates occur before agent response.
- Includes scripts for classification, decay, session cache, health checks, and secure compaction/extraction of conversation history.
- Supports both keyword- and LLM-based memory extraction/compaction modes for minimal or nuanced sessions.
元数据
常见问题
Smart Memory Plus 是什么？
Complete memory system for OpenClaw agents. Combines enhanced memory management (WAL protocol, type classification, temporal decay, session cache) with conte... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 157 次。
如何安装 Smart Memory Plus？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install smart-memory-plus」即可一键安装，无需额外配置。
Smart Memory Plus 是免费的吗？
是的，Smart Memory Plus 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
Smart Memory Plus 支持哪些平台？
Smart Memory Plus 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 Smart Memory Plus？
由 zgjq（@zgjq）开发并维护，当前版本 v1.4.0。
推荐 Skills