The community hub for tools, skills, and infrastructure extensions for OpenClaw — the self-hosted AI agent platform.
OpenClaw gives you a personal AI agent that lives on your own machine: no subscriptions, no vendor lock-in, no data leaving your infrastructure. This repo is the index of everything the community has built on top of it — from OS-level daemons that keep your agent healthy 24/7, to skills that turn your agent into a full engineering team.
Everything here is open source, agent-operable, and designed around one rule: if your agent can't run it without you, it doesn't belong here.
Core guarantees across every entry:
- ✅ One-liner install
- ✅ Works on macOS and Linux
- ✅ Human-readable README + agent-readable
SKILL.mdor runbook in every repo - ✅ No mandatory cloud services
New to OpenClaw? Start here.
1. Install OpenClaw
npm install -g openclaw
openclaw gateway startFull docs at github.com/openclaw/openclaw.
2. Pick your infrastructure layer (tools below) — these run at the OS level alongside your gateway and keep it healthy, observable, and backed up.
3. Drop in skills — these live in ~/.openclaw/workspace/skills/ and your agent discovers and uses them automatically when tasks match.
4. Talk to your agent. It knows about the tools it has. You don't need to invoke skills by name.
These run at the OS level alongside OpenClaw — as launchd services (macOS) or systemd units (Linux). Install them once and forget they exist until you need them.
| Tool | Problem it solves | Install |
|---|---|---|
| openclaw-sentinel | Multiple watchdog layers make recovery unpredictable. Sentinel keeps a single recovery loop: health probe, rollback, agent repair, and Telegram rescue without overlapping daemons. | curl -fsSL https://raw.githubusercontent.com/bkochavy/openclaw-sentinel/main/install.sh | bash |
| openclaw-backup | One bad openclaw.json edit and your config is gone. Daily encrypted backup to a private GitHub repo, with a verified push check and a Telegram alert if anything fails — before you even notice a problem. |
curl -fsSL https://raw.githubusercontent.com/bkochavy/openclaw-backup/main/install.sh | bash |
Archived / legacy:
openclaw-watchdogis legacy.openclaw-sentinelis the active v2 recovery path.
These install through openclaw plugins install and are meant to stay small, explicit, and easy to audit.
| Plugin | Problem it solves | Install |
|---|---|---|
| openclaw-response-validator | The agent did the tool work but never actually answered. Detects orphaned tool-call endings and can nudge the next turn to close the loop. | curl -fsSL https://raw.githubusercontent.com/bkochavy/openclaw-response-validator/main/install.sh | bash |
| openclaw-tool-result-guard | Huge tool payloads make transcripts heavy and follow-up replies sticky. Truncates oversized tool_result payloads before persistence. |
curl -fsSL https://raw.githubusercontent.com/bkochavy/openclaw-tool-result-guard/main/install.sh | bash |
| openclaw-postfix-pack | You want compact model stamps without patching upstream. V2 uses a native responsePrefix marker plus a tiny managed plugin to append the real provider/model suffix. |
curl -fsSL https://raw.githubusercontent.com/bkochavy/openclaw-postfix-pack/main/install.sh | bash |
Drop these into ~/.openclaw/workspace/skills/ and your agent picks them up immediately — no restart required. Each skill ships with a SKILL.md that your agent reads to understand when and how to use it.
| Skill | Problem it solves | Install |
|---|---|---|
| openclaw-tailserve | Sharing a local file or port always takes too many steps. TailServe wraps Tailscale to give your agent a single command that produces a verified HTTPS URL — shareable in Telegram, browser, or anywhere else — in under a second. | npm install -g tailserve |
| openclaw-build | Going from idea to shipped code still requires you in the loop. Build runs a PRD interview, gets your approval on the spec, then hands off to a coding loop (Codex or Claude Code) and doesn't stop until it's done and tested. | git clone https://github.com/bkochavy/openclaw-build ~/.openclaw/workspace/skills/openclaw-build |
| openclaw-agent-dashboard | You have no idea what your sub-agents are actually doing. The dashboard gives you a live browser view of every running tmux and coding session — task description, git status, token usage, and whether it's spinning or stuck. | git clone https://github.com/bkochavy/openclaw-agent-dashboard ~/.openclaw/workspace/skills/openclaw-agent-dashboard && node ~/.openclaw/workspace/skills/openclaw-agent-dashboard/server.js |
Every tool and skill here follows the same design contract:
For humans: one-sentence problem statement, one-line install, plain English docs. You should be able to evaluate, install, and trust a tool in under 5 minutes.
For agents: a SKILL.md or agent-readable runbook section with exact commands, expected outputs, exit codes, and failure handling — so your OpenClaw agent can operate these tools autonomously without interrupting you.
Cross-platform: macOS (launchd) and Linux (systemd). No Homebrew dependency for core functionality. No cloud account required to get value.
Open a PR or issue with your tool or skill. Requirements:
- Open source (MIT or Apache 2.0)
- Works on a fresh OpenClaw install (macOS or Linux)
- Has both a human-facing README section and an agent-facing
SKILL.mdor runbook - No hardcoded personal data, credentials, or user-specific paths
- Tested on at least one real OpenClaw deployment
We're not trying to be exhaustive — we're trying to be trustworthy. Quality over quantity.
- OpenClaw — the platform these tools extend