Skill flagged — suspicious patterns detected

ClawHub Security flagged this skill as suspicious. Review the scan results before using.

Shared Memory Stack

v1.0.0

Complete reference for the shared memory architecture connecting Claude Code, OpenClaw/Kimi, and LM Studio subagents through Obsidian vault + MemPalace (Chro...

0· 70·0 current·0 all-time

Install

OpenClaw Prompt Flow

Install with OpenClaw

Best for remote or guided setup. Copy the exact prompt, then paste it into OpenClaw for nerua1/shared-memory-stack.

Previewing Install & Setup.
Prompt PreviewInstall & Setup
Install the skill "Shared Memory Stack" (nerua1/shared-memory-stack) from ClawHub.
Skill page: https://clawhub.ai/nerua1/shared-memory-stack
Keep the work scoped to this skill only.
After install, inspect the skill metadata and help me finish setup.
Use only the metadata you can verify from ClawHub; do not invent missing requirements.
Ask before making any broader environment changes.

Command Line

CLI Commands

Use the direct CLI path if you want to install manually and keep every step visible.

OpenClaw CLI

Bare skill slug

openclaw skills install shared-memory-stack

ClawHub CLI

Package manager switcher

npx clawhub@latest install shared-memory-stack
Security Scan
Capability signals
CryptoRequires OAuth token
These labels describe what authority the skill may exercise. They are separate from suspicious or malicious moderation verdicts.
VirusTotalVirusTotal
Suspicious
View report →
OpenClawOpenClaw
Suspicious
medium confidence
!
Purpose & Capability
The documented purpose — coordinating a shared local memory (Obsidian vault + ChromaDB/MemPalace) between Claude Code, OpenClaw and LM Studio — is coherent with the instructions. However the registry metadata declares no required binaries, env vars, or config paths while the SKILL.md repeatedly references specific binaries (/opt/homebrew/bin/mempalace, /opt/homebrew/bin/capture-idea, openclaw CLI, python3), absolute filesystem paths (/Volumes/2TB_APFS/...), and GitHub/SSH usage. That mismatch (metadata says 'none' but the doc requires many local tools/keys) is an incoherence and should be explained by the author before use.
!
Instruction Scope
The SKILL.md tells operators to read/write specific local directories, run local binaries that mine and index vault contents, and communicate over a local gateway (port 18789). It also references publishing to GitHub and assumes an SSH key and gh are configured. These instructions require filesystem access and existing credentials; if an agent were allowed to execute them automatically they could read and transmit potentially sensitive local data. The instructions are not purely descriptive — they include concrete commands that would perform IO on the host.
Install Mechanism
No install specification or code files are present; the skill is instruction-only. This reduces supply-chain risk because nothing will be automatically downloaded or written by the registry install itself. All executable behavior depends on local binaries described in the documentation.
!
Credentials
The skill metadata lists no required environment variables or credentials, yet the documentation explicitly assumes access to an SSH key and the 'gh' CLI, a local OpenClaw gateway, and specific home/workspace directories. That gap is concerning: the skill expects privileged local artifacts (keys, repos, large data directories) but does not declare them. Users should not expose SSH keys, tokens, or allow automatic execution of the described commands without verifying what will run and what data will be read or transmitted.
Persistence & Privilege
The skill is not force-included (always: false) and is user-invocable. Autonomous invocation is permitted by default but is not by itself flagged here. The skill does not request to modify other skills or system-wide settings in the provided documentation.
Scan Findings in Context
[no_regex_findings] expected: The static regex scanner had no code files to analyze (instruction-only). Absence of code findings is expected for a SKILL.md-only skill, but does not imply the instructions are safe to execute.
What to consider before installing
This skill is a documentation-only guide for running a local shared-memory stack; it appears coherent in purpose but the metadata omits many operational requirements. Before installing or running anything mentioned here: 1) Do not allow an agent to run these commands autonomously — the doc includes concrete commands that will read and write local directories and could transmit data. 2) Manually verify the existence and contents of the referenced paths and binaries (/Volumes/2TB_APFS/..., /opt/homebrew/bin/mempalace, capture-idea, openclaw CLI). 3) Confirm SSH keys and 'gh' config are present and intended for use; never expose or copy private keys. 4) Review any local scripts (capture-idea, mempalace wrapper) before execution to ensure they don't call out to remote endpoints or exfiltrate data. 5) Ask the skill author to update metadata to list required binaries/credentials and to clarify which actions are purely descriptive vs meant to be executed. If you cannot validate these points, treat the skill as high-risk and avoid running the documented commands.

Like a lobster shell, security has layers — review code before you run it.

latestvk97dq8j20e1f6ky9nktrzcgpc984pb7k
70downloads
0stars
1versions
Updated 2w ago
v1.0.0
MIT-0

shared-memory-stack

Shared memory architecture for a multi-agent system: Claude Code + OpenClaw/Kimi + LM Studio. No idea gets lost. All agents read and write the same memory.

Architecture overview

┌─────────────────────────────────────────────────────┐
│                    SHARED MEMORY                     │
│                                                      │
│  Obsidian vault (structural)                         │
│  /Volumes/2TB_APFS/openclaw-data/workspace/          │
│    obsidian-memory/                                  │
│    ├── wiki/        ← MOCs, documentation            │
│    ├── daily/       ← session logs, daily notes      │
│    ├── answers/     ← resolved Q&A                   │
│    ├── source/      ← raw research, clips            │
│    └── ideas/       ← captured ideas (searchable)   │
│                                                      │
│  MemPalace (semantic / ChromaDB)                     │
│  /Volumes/2TB_APFS/openclaw-data/workspace/          │
│    memory/palace/   ← vector index (25 drawers)     │
└─────────────────────────────────────────────────────┘
         ▲                        ▲
         │ reads/writes           │ mines vault
  ┌──────┴──────┐         ┌───────┴───────┐
  │  OpenClaw   │◄───────►│  Claude Code  │
  │  (Kimi)     │ bridge  │  (this agent) │
  └──────┬──────┘         └───────────────┘
         │
  ┌──────┴──────┐
  │  LM Studio  │
  │  subagents  │
  └─────────────┘

Key paths

WhatPath
Obsidian vault/Volumes/2TB_APFS/openclaw-data/workspace/obsidian-memory/
Ideas dir/Volumes/2TB_APFS/openclaw-data/workspace/obsidian-memory/ideas/
MemPalace palace/Volumes/2TB_APFS/openclaw-data/workspace/memory/palace/
mempalace.yaml/Volumes/2TB_APFS/openclaw-data/workspace/obsidian-memory/mempalace.yaml
Python 3.12 venv/Volumes/2TB_APFS/openclaw-data/workspace/memory/mempalace-venv/
capture-idea script/Volumes/2TB_APFS/openclaw-data/workspace/scripts/capture-idea.sh
capture-idea binary/opt/homebrew/bin/capture-idea
OpenClaw skills/Volumes/2TB_APFS/openclaw-data/workspace/skills/
Claude Code skills~/.claude/skills/vault/

1. MemPalace setup

Why Python 3.12 venv

macOS ships Python 3.14 which breaks chromadb (pydantic v1 incompatibility). MemPalace runs in a dedicated Python 3.12 venv.

/opt/homebrew/bin/mempalace   ← wrapper (zsh)
  → VENV/bin/mempalace --palace PALACE "$@"

The wrapper auto-injects --palace so no flag needed in daily use.

Mining vault into palace

# Mine entire vault (run after adding new files)
mempalace mine /Volumes/2TB_APFS/openclaw-data/workspace/obsidian-memory/ --wing obsidian_memory

# Dry run first
mempalace mine ... --dry-run

# Status
mempalace status

Searching

# Semantic search
mempalace search "oauth token expiry mobile"

# Search by topic/tag (frontmatter grep)
grep -r "topic: bezpieczenstwo" obsidian-memory/ideas/ -l
grep -r "tags:.*oauth" obsidian-memory/ideas/ -l

# Search by date range
ls obsidian-memory/ideas/2026-04-*.md

Rooms

RoomSource dirKeywords
documentationwiki/documentation, wiki
dailydaily/daily
ideasideas/idea, pomysl, insight, problem
answersanswers/answers
general(fallback)

2. Capture pipeline

Every valuable idea, observation, or problem from any agent session → saved to vault → indexed in MemPalace.

From CLI (any agent)

capture-idea \
  --title "OAuth token expiry zbyt krótki na mobile" \
  --topic bezpieczenstwo \
  --tags "oauth,token,mobile,auth" \
  --body "Tokeny wygasają po 1h, użytkownicy mobilni są wylogowywani." \
  --source openclaw   # or: claude-code | lmstudio | manual

Simple mode:

capture-idea "Krótki opis idei"

From Claude Code

/capture problem z oauth tokenami wygasającymi za szybko na mobile

Idea file format

---
date: 2026-04-12
source: openclaw
topic: bezpieczenstwo
tags: [oauth, token, mobile]
related: []
status: seedling
---

# Tytuł

Treść...

Topic values: architektura | bezpieczenstwo | ux | performance | integracja | ai | devops | dane | inne

Status values: seedlinggrowingmature

Files land in: obsidian-memory/ideas/YYYY-MM-DD-slug.md Auto-mined into MemPalace after each capture.


3. Inter-agent communication

Claude Code and OpenClaw communicate via the local OpenClaw gateway (port 18789).

Claude Code → OpenClaw

openclaw agent --message "Twoja wiadomość" --agent main --json

Parse response:

openclaw agent --message "..." --agent main --json | python3 -c "
import json, sys
d = json.load(sys.stdin)
for p in d['result']['payloads']:
    if p.get('text'): print(p['text'])
"

Claude Code skill: openclaw-bridge

Load with /skill openclaw-bridge or invoke directly:

/ask-openclaw Czy ta migracja SQL jest bezpieczna?

OpenClaw → Claude Code

OpenClaw writes to shared vault/files. Claude Code reads on next session via memory files at: ~/.claude/projects/-Volumes-2TB-APFS/memory/


4. Publishing skills

Both agents publish to GitHub under nerua1. SSH key and gh are configured.

# Claude Code skills
cd ~/.claude/skills/vault/SKILL_NAME
git init && git add . && git commit -m "feat: SKILL_NAME v1.0.0"
gh repo create SKILL_NAME --public --source . --remote origin --push

# OpenClaw skills
/publish-skill SKILL_NAME

Configured:

  • SSH key: ~/.ssh/github_nerua1 (added to github.com/nerua1)
  • gh authenticated as nerua1
  • git config: user.name=nerua1, user.email=neru_a1@icloud.com

5. Quick reference

# Health check
mempalace status
openclaw health

# Mine vault after changes
mempalace mine /Volumes/2TB_APFS/openclaw-data/workspace/obsidian-memory/ --wing obsidian_memory

# Capture idea
capture-idea --title "..." --topic ai --tags "..." --body "..."

# Search memory
mempalace search "keyword"

# Ask OpenClaw
openclaw agent --message "..." --agent main --json

# Publish a skill
gh repo create SKILL_NAME --public --source . --remote origin --push

6. Known constraints

IssueWorkaround
Python 3.14 breaks chromadbUse /opt/homebrew/bin/mempalace wrapper (Python 3.12 venv)
mempalace skips symlinksMine real vault path, not sources/obsidian/ symlink dir
OpenClaw gateway loopback onlyCommunication only works on same machine
Claude Code has no persistent processShared state via filesystem only

Comments

Loading comments...