agent-pack-n-go
🚀 Clone your OpenClaw AI Agent to a new device in ~25 minutes — configs, memory, skills, everything.
Install, harden, and run the MoltWorld Dashboard reliably for real users. Use when asked to set up local runtime scaffolding (README/package.json/.env/.gitignore), add Docker/Compose/systemd deployment files, verify accessibility on port 8787, and troubleshoot uptime/connectivity issues.
Suivez les instructions du dépôt pour votre agent. Consultez les fichiers et dépendances avant l’installation.
SkillHub CLI
skillhub install guillaumetch/moltworld-dashboard-deployExaminez la source et la compatibilité, puis suivez les étapes d’installation.
Donnez le lien à votre agent et demandez-lui d’examiner instructions et dépendances avant d’expliquer les étapes.
Standardize this workflow to make moltworld-dashboard easy to run and operate safely.
Required binaries:
Optional (privileged persistence path only):
package.json and lockfile for suspicious scripts/deps.sudo, systemctl enable, writing under /etc/systemd/system).curl | bash or equivalent remote script execution.server.mjs, public/).package.json (start script).env.example
.nvmrc
.gitignore
README.md
Dockerfile
docker-compose.yml
moltworld-dashboard.service (systemd, only with explicit approval)http://localhost:8787/.data/state.json, logs, pids)..env).8787.Use these checks when service becomes unreachable:
ss -ltnp | grep ':8787' || true
curl -I --max-time 5 http://localhost:8787/
If process is down, restart with a supervisor (Docker Compose or approved systemd service) instead of ad-hoc foreground runs.
Ces informations proviennent de l’analyse indexée et peuvent être anciennes. Vérifiez les fichiers actuels.
# MoltWorld Dashboard Deploy Standardize this workflow to make `moltworld-dashboard` easy to run and operate safely. ## Runtime requirements Required binaries: - npm - docker - docker-compose Optional (privileged persistence path only): - systemd - sudo ## Safety gates (mandatory) 1. Treat project dependencies as untrusted until reviewed. 2. Before package install, inspect `package.json` and lockfile for suspicious scripts/deps. 3. Ask for explicit approval before any privileged/persistent action (`sudo`, `systemctl enable`, writing under `/etc/systemd/system`). 4. Prefer non-privileged runtime paths first (local run or Docker Compose without host-level service install). 5. Never use `curl | bash` or equivalent remote script execution. ## Workflow 1. Verify baseline project files exist (`server.mjs`, `public/`). 2. Add/shareability files if missing: - `package.json` (start script) - `.env.example` - `.nvmrc` - `.gitignore` - `README.md` 3. Add deployment files if requested: - `Dockerfile` - `docker-compose.yml` - `moltworld-dashboard.service` (systemd, only with explicit approval) 4. Validate startup and confirm HTTP 200 on `http://localhost:8787/`. 5. Validate restart behavior and long-running stability. 6. Confirm accessibility via localhost or host IP. 7. Document runbook steps for operators. ## Required file conventions - Keep runtime state out of git (`data/state.json`, logs, pids). - Keep secrets out of git (`.env`). - Default runtime port: `8787`. - README must include: - local quick start - Docker run - Docker Compose run - systemd install/enable instructions marked as privileged/optional ## Runtime stability checks Use these checks when service becomes unreachable: ```bash ss -ltnp | grep ':8787' || true curl -I --max-time 5 http://localhost:8787/ ``` If process is down, restart with a supervisor (Docker Compose or approved systemd service) instead of ad-hoc foreground runs. ## Troubleshooting quick checks - S...
🚀 Clone your OpenClaw AI Agent to a new device in ~25 minutes — configs, memory, skills, everything.
⚡ SOTA agent skills for OpenClaw — observability, security, code quality, incident response, and more. Built by Anvil AI.
OpenClaw Skill: Safely restart the Gateway with context preservation, guardian watchdog, and multi-channel notification
Automated Linux server patching with PatchMon integration for OpenClaw