We build hands for AI that moves first.
effectorHQ (prev. OpenClawHQ) is an open studio building the next generation of Proactive AI agent capabilities. Each project aims to solve a real problem, is backed by real research, and is designed to stand on its own. Together they form a coherent vision: typed, composable, verifiable hands for AI.
Starting with OpenClaw, designed for a larger capability layer. We exist because one team can't cover every use case. Distributed, effector-native, community-driven is a better architecture.
We don't build the brain (that's the LLM). We don't build the body (that's the runtime — OpenClaw, Claude Agent SDK, and whatever comes next). We build the hands — and we make sure every hand is safer, more composable, and more powerful than what existed before.
graph LR
classDef brain fill:#1A1A1A,color:#F5F0EB,stroke:#1A1A1A;
classDef body fill:#2D2D2D,color:#F5F0EB,stroke:#1A1A1A;
classDef hands fill:#E03E3E,color:#F5F0EB,stroke:#E03E3E;
Brain["Brain / LLMs"]:::brain --> Body["Body / Runtimes"]:::body
Body --> Hands["Hands / Effectors"]:::hands
Our analysis of 13,729 ClawHub skills found a 67% agent failure rate: driven by untyped interfaces, missing prerequisites, and permission mismatches.
This is the paradigm we're building: Effector = the smallest capability unit an agent can discover, inspect, run, compose, and carry across runtimes.
The core thesis: AI agent capabilities need types. Today you chain two skills and pray they work. Tomorrow you type-check the composition before a single token is spent.
| Project | What it is | Status |
|---|---|---|
effector-spec |
The Specification — type language, composition algebra, discovery protocol | v0.2.0 draft |
effector-types |
Standard capability types — the lib.d.ts for agent tools, grounded in 13K+ skills |
v0.2.0 |
clawhub-analysis |
Empirical data — 13,729 skills analyzed, type distributions, failure rates | Data + notebook |
what makes capabilities reliable
| Project | What it does | Status |
|---|---|---|
skill-lint |
CLI — validate SKILL.md structure before publishing | Working (v0.2.0) |
skill-eval |
Evaluation framework — measure whether skills actually work | v0.1.0 (static) |
effector-audit |
Security audit — Sigstore signing, permission drift detection | v0.1.0 |
skill-lint-action |
GitHub Action — inline PR annotations, zero config | Working |
get started in minutes
| Project | What it does | Status |
|---|---|---|
create-effector |
npx create-effector — scaffold any Effector type |
Working |
linear-skill |
Reference implementation — production-ready skill, Grade A eval, zero lint errors | v1.0.0 |
plugin-template |
Skill starter template — fork and write your SKILL.md | Working |
cookbook |
Real-world skill recipes: Docker, Jira, PostgreSQL, git worktrees | Examples |
connecting capabilities
| Project | What it does | Status |
|---|---|---|
effector-compose |
Build agent pipelines, type-check them, emit to any runtime | Early |
effector-graph |
Capability graph visualization and path-finding | Early |
openclaw-mcp |
SKILL.md → MCP bridge — make skills work in Claude, Cursor, Windsurf | v0.1.0 |
| Project | What it does |
|---|---|
awesome-openclaw |
Curated list of skills, extensions, tools, and resources |
workspace-templates |
Agent persona bundles (DevOps, Code Reviewer, Security Auditor) |
lobster-recipes |
Workflow pipelines: deploy-and-notify, daily-standup, PR-review-triage |
docs |
Community guides: architecture, skill development, extension development |
| Project | What it does |
|---|---|
OpentheClaw |
One‑click SSH tunnel + Web login |
free-search-proxy |
DuckDuckGo‑based search proxy API for OpenClaw agents |
| Document | Purpose |
|---|---|
manifesto |
Founding theses — why typed capabilities matter |
rfcs |
RFC process for spec changes and new products |
| Contributing Guide | How to contribute to any project |
| Effector Type | What it is |
|---|---|
| Skill Effectors | Markdown-defined capabilities (SKILL.md) with typed interfaces |
| Extension Effectors | TypeScript plugins that hook into runtime SDKs |
| Workflow Effectors | Multi-step pipelines with typed composition |
| Workspace Effectors | Agent persona bundles (SOUL.md, AGENTS.md, TOOLS.md) |
| Bridge Effectors | Cross-runtime adapters (OpenClaw ↔ MCP ↔ others) |
| Quality Effectors | Lint, eval, audit, and supply-chain tools |
| Principle | In practice |
|---|---|
| Demo First | Ship working code before writing proposals. A running prototype beats a perfect spec. |
| Ship Loud | Every merge, every release note, every contributor gets signal. Shipping should feel like winning. |
| Open by Default | Code, roadmap, decisions, failed experiments — if it can be open, it is. |
This is a community project. No gatekeeping. Browse a repo, open a PR, or start something new.
If you have questions, open an issue in the relevant repo or start a Discussion.
Every effector extends the reach. © 2026 effectorHQ Contributors
