← Back to Skills Marketplace
688
Downloads
0
Stars
4
Active Installs
3
Versions
Install in OpenClaw
/install memory-pro
Description
This skill provides semantic search over your memory files using a local vector database.
Usage Guidance
This package appears to implement a legitimate local semantic-search service, but it contains several red flags you should evaluate before installing or running it: 1) The manifest marks many environment variables as "required" (and calls them credentials) even though most are optional tuning flags — don't populate secrets blindly. 2) If you enable reranking (MEMORY_PRO_ENABLE_RERANK or set RERANK_PROVIDER/ENDPOINT/API_KEY), the service will send candidate sentences (your memory text) to the configured external endpoint — treat that as potential data exfiltration. 3) The code reads files across your workspace (memory markdowns and core files) and will write indexes and tmp files; review those paths and remove any bundled state before use. 4) Validation scripts try to restart a user systemd service (systemctl --user restart), which modifies system state — avoid running validation scripts that alter services unless you trust and inspected them. Recommended actions: run this in an isolated environment (container or throwaway VM), create a minimal .env that points to a test memory directory (not your real workspace), do not set rerank-related env vars or external endpoints unless you understand the data flows, and review/limit the configured MEMORY_PRO_DATA_DIR and core file list before building the index. If you need this skill but lack confidence, ask the publisher to: a) mark only truly required env vars as required, b) document rerank behavior clearly, and c) avoid requiring systemctl in validation scripts.
Capability Analysis
Type: OpenClaw Skill
Name: memory-pro
Version: 2.5.0
The bundle implements a sophisticated local RAG (Retrieval-Augmented Generation) system for semantic search over a user's workspace. It features a FastAPI-based service (v2/main.py), FAISS vector indexing (v2/build_index.py), and a hybrid retrieval engine (v2/retrieval_hybrid.py) that combines vector search with BM25 and recency scoring. While v2/rerank.py includes functionality to send text snippets to external APIs (like Jina AI), this is a standard industry practice for RAG systems and is disabled by default, requiring explicit user configuration via environment variables. The use of systemd for persistence and broad workspace access is consistent with the tool's stated purpose of providing a continuous background indexing service.
Capability Assessment
Purpose & Capability
Name/description (local semantic search over memory files) match the included code (indexing, vector search, hybrid retrieval, CLI client). However the manifest marks many MEMORY_PRO_* and OPENCLAW_* environment variables as required (and labels them all as "Credential used by memory-pro") even though many are optional tuning/config values (rerank provider, rerank API key, ports, weights, flags). Marking HOME as the primary credential is incorrect: HOME is a path, not a secret credential. The large list of required env vars is disproportionate to a minimal local memory-search skill.
Instruction Scope
Runtime code and scripts read many user workspace files (memory markdowns, core files like MEMORY.md/AGENTS.md/USER.md), rebuild local FAISS indexes, and expose a local HTTP API. More importantly, the rerank path can post candidate documents (user memory sentences) to external rerank endpoints (e.g., Jina or an OpenAI-compatible endpoint) if reranking is enabled and configured—this allows exfiltration of indexed content. Validation script (v2/validate_phase1.sh) calls 'systemctl --user restart memory-pro.service', which tries to control a systemd user service (a scope/privilege escalation beyond simple local indexing).
Install Mechanism
No install spec is provided (instruction-only). The skill ships Python scripts and shell helpers but does not download arbitrary binaries from remote URLs during install. This is lower risk from an installation perspective.
Credentials
The manifest declares a very large number of required environment variables (many are tuning flags or optional: rerank provider/endpoint/API key/model, MMR flags, BM25 paths, etc.). Several of these (RERANK_API_KEY, RERANK_ENDPOINT, RERANK_PROVIDER) enable sending user data to external services if set—yet they are listed as required. Declaring HOME as the primary credential is incorrect and misleading. Required config paths include files under /skills/... and /tmp which increases the surface of files the skill expects to access.
Persistence & Privilege
The skill does not request always:true, and normally does not grant extra autonomous privileges — that is fine. However the validation script attempts to restart a user systemd service (systemctl --user restart memory-pro.service) which modifies system-level state. The start script also rebuilds the index and starts a local HTTP server (uvicorn) — normal for a service but it means the skill will write files (index, sentences.txt, bm25 payload) to disk and listen on a port. Combined with the rerank/external-call path, that increases the blast radius.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install memory-pro - After installation, invoke the skill by name or use
/memory-pro - Provide required inputs per the skill's parameter spec and get structured output
Version History
v2.5.0
**Major update: Overhauled structure with new v2 engine and improved configuration.**
- updated to support openclaw 2026.3.11 or later
- Migrated core logic to a new `v2` directory with updated engine scripts and benchmarks.
- Replaced old `engine/` files and installation scripts with new modular scripts for building, searching, and benchmarking.
- Updated and expanded environment variable and configuration file support for detailed customization.
- SKILL.md and documentation revised for v2 architecture, service setup, and usage.
- Adds scripts for semantic search, hybrid retrieval, reranking, and evaluation.
- Removes deprecated files and legacy install process.
v2.0.1
- Added SECURITY.md to document security-related policies.
- Removed .gitignore and README.md files.
- No changes to core skill logic or documentation in SKILL.md.
v2.0.0
All notable changes to this project will be documented in this file.
The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
## [2.0.0] - 2026-02-22
### Changed
- **Architectural Shift**: Transitioned from a legacy keyword-based search to a highly efficient **Semantic Search** engine powered by `all-MiniLM-L6-v2` and FAISS.
- **Agent Integration**: Client script (`search.py`) is now completely decoupled and dependency-free (using pure Python `urllib` instead of `requests`), allowing OpenClaw agents to invoke it seamlessly across different hosts.
- **Path Resolution**: Removed all hardcoded absolute paths. Replaced with robust Environment Variable mapping (`MEMORY_PRO_WORKSPACE_DIR`, `MEMORY_PRO_DATA_DIR`) for zero-configuration portability.
- **Background Service**: Replaced foreground script execution with a dedicated `systemd` user service (`memory-pro.service`) that natively manages the FastAPI lifecycle.
- **Index Management**: Added an automatic index consistency check (`build_index.py`) that runs before the FastAPI server starts, ensuring the FAISS `.index` file and text mappings are always synchronized.
### Added
- Automated `install.sh` script to cleanly clone, configure, and initialize the Python virtual environment (via `uv`) and systemd service.
- Robust JSON output flag (`--json`) to `search.py` for structured tool parsing by AI agents.
- Automatic retry and backoff logic inside the client script to handle service restarts gracefully.
- Healthcheck endpoint (`/health`) to monitor initialization state and indexed item count.
### Removed
- Removed old keyword dependency logic.
- Removed hard coupling to the user's specific NAS/Linux setup, making the skill globally applicable for any OpenClaw user.
Metadata
Frequently Asked Questions
What is memory-pro?
This skill provides semantic search over your memory files using a local vector database. It is an AI Agent Skill for Claude Code / OpenClaw, with 688 downloads so far.
How do I install memory-pro?
Run "/install memory-pro" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is memory-pro free?
Yes, memory-pro is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does memory-pro support?
memory-pro is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created memory-pro?
It is built and maintained by royhk920 (@royhk920); the current version is v2.5.0.
More Skills