1Claw Agent Templates

Public template registry for 1claw spawn — framework-specific Docker agents pre-wired with 1Claw MCP and Shroud.

Each template includes a Dockerfile, starter agent code, and an entrypoint that mounts the host daemon socket. A shared browser chat UI on port 3000 lets you test any template immediately after 1claw spawn — messages route to the framework agent (port 3001) or Shroud via the daemon when no LLM is wired in the agent. The container never sees raw API keys; credentials are injected at runtime by the 1Claw daemon on the host.

Full contributor guide (docs): docs.1claw.co — Add an agent template

---

Use a template

Install the CLI, then spawn:

npm install -g @1claw/cli   # or: brew install 1clawAI/tap/oneclaw

1claw spawn langchain --agent-key ocv_YOUR_KEY
1claw spawn --list
1claw spawn --refresh     # pull latest templates from this repo

See the CLI guide for LLM options (BYOK, Token Billing, offline mode).

v0.56 platform features (configure on agents)

Spawned agents use the 1Claw API. Enable on agent records via dashboard or CLI:

• Graduated HITL — Transactions, sign intents, execution bindings may require human approval.
• Human Factor Auth — Treasury send/swap passkey step-up (human wallets; @1claw/wallet-react).
• Guardrail governance — Shadow/enforce execution guardrails, revision history, replay dry-runs.
• Safe foundation — Counterfactual Gnosis Safe accounts (1claw agent accounts migrate).
• Multichain signing — Unchanged deps in Vault/Shroud: rust-bitcoin, solana-sdk v4, xrpl-rust 1.1.0.

v0.42 features available in all templates

All Python templates now include the oneclaw SDK (v0.42+), giving spawned agents access to:

• Agent Memory — client.memory.store() / client.memory.search() for persistent vector memory across sessions
• Automations — client.automations.create() for cron-based scheduled tasks (secret rotation, health checks)
• Runtimes — client.runtimes.create() for managed deployment with auto-scaling
• Discovery — client.discovery.publish() to list your agent in the 1Claw directory

These complement the MCP-based secret fetching and Shroud LLM proxy already wired in each template.

Shipped templates

Python:

Template	Framework	Status
langchain	LangChain / LangGraph	✅
crewai	CrewAI	✅
openai-agents	OpenAI Agents SDK	✅
agentkit	Coinbase AgentKit	✅
smolagents	HuggingFace smolagents	✅
llamaindex	LlamaIndex	✅
pydantic-ai	Pydantic AI	✅
agno	Agno	✅
coder	Coder	✅

TypeScript:

Template	Framework	Status
typescript-sdk	@1claw/sdk + AI SDK	✅
mastra	Mastra	✅
elizaos	ElizaOS	✅

---

Contribute a template

We welcome community PRs for new frameworks and improvements to existing templates.

Before you start

• Pick a unique directory name (lowercase, hyphens): e.g. pydantic-ai, my-framework.
• Copy an existing template as a starting point:
  • Python: templates/langchain/
  • TypeScript: templates/typescript-sdk/ or templates/mastra/
• Read the docs contributor guide for the full walkthrough.

Step-by-step

1. Fork 1clawAI/agent-templates and clone your fork.

2. Create templates/<name>/ with these files:

   File	Required
   template.yaml	Manifest (name must match directory)
   Dockerfile	Builds a runnable image
   entrypoint.sh	Starts MCP + agent; wires Shroud when enabled
   Starter code	e.g. agent.py or agent.ts
   README.md	Framework-specific notes
   requirements.txt	Python deps (Python templates)
   package.json	Node deps (TypeScript templates)

3. Register the template in registry.yaml:

   - name: my-framework          # must match templates/my-framework/
     display_name: "My Framework"
     version: 1.0.0
     language: python             # python | node
     description: "One-line summary for 1claw spawn --list"

4. Test locally (build from repo root so shared/ is in the Docker context):

   docker build -f templates/my-framework/Dockerfile -t test-my-framework .
   docker run --rm -p 3000:3000 test-my-framework
   open http://localhost:3000    # interactive chat UI
   curl http://localhost:3000/health

   Optionally test end-to-end with the CLI (from a monorepo checkout with this repo at packages/agent-templates):

   1claw spawn my-framework --agent-key ocv_YOUR_KEY

5. Open a pull request against main. CI will:

   • Validate template.yaml (required fields, name match, language)
   • Confirm the template is listed in registry.yaml
   • Scan for accidental secrets (sk-…, ocv_…, etc.)
   • Run docker build smoke tests (see validate.yml)

6. After merge, templates appear for everyone via 1claw spawn --refresh (or on the next CLI release that bundles this repo).

Rules (must follow)

• No secrets in the repo. Never commit API keys, agent keys, or real vault IDs.
• Daemon socket, not env secrets. Templates must work with ONECLAW_DAEMON_SOCKET=/run/1claw/daemon.sock; do not ENV or COPY credentials.
• Health endpoint. Expose /health on port 3000 via the shared spawn chat UI (shared/start-with-chat-ui.sh).
• LLM via Shroud. Route provider calls through Shroud when ONECLAW_LLM_VIA_SHROUD=true (see entrypoint.sh in existing templates). Use the daemon /proxy path in the shared chat UI fallback — never put real keys in the container.

Deep reference

• CONTRIBUTING.md — manifest schema, Dockerfile guidelines, entrypoint pattern
• Docs: Add an agent template — step-by-step guide with examples
• CLI spawn docs — user-facing 1claw spawn reference

Questions?

Open a GitHub Discussion or issue. For security concerns, email ops@1claw.co.

---

Repository layout

agent-templates/
├── registry.yaml           # Index consumed by 1claw spawn --list
├── shared/                 # Spawn chat UI (browser test harness for all templates)
│   ├── spawn_chat_ui.py
│   ├── index.html
│   └── start-with-chat-ui.sh
├── templates/
│   ├── langchain/          # One directory per framework
│   │   ├── template.yaml
│   │   ├── Dockerfile
│   │   ├── entrypoint.sh
│   │   └── ...
│   └── ...
├── CONTRIBUTING.md         # Technical schema & patterns
└── .github/workflows/      # CI validation

License

MIT — see LICENSE.