Skip to content

setup.md

Latest commit

 

History

History
111 lines (85 loc) · 5.99 KB

setup.md

File metadata and controls

111 lines (85 loc) · 5.99 KB
Raw

Setup — prerequisites, then let an agent do the rest

Setting up this repo is deliberately short: you handle the accounts and installs no software can do for you, then an agent runs the actual wiring — the fleet-setup skill is the executable version of this guide.

Works on Linux, macOS, and Windows. The shell scripts are Bash: on Windows, run them inside WSL (recommended) or Git Bash.

1. Prerequisites — the only manual part

What How Check
Clawdi account sign up and log in at clawdi.ai dashboard loads
Node.js ≥ 20, clawdi + foc-cli CLIs npm i -g clawdi foc-cli → clawdi auth login clawdi auth status --json → "authenticated": true (parse the field — the exit code is 0 even when logged out)
Hosted OpenClaw agent create it in the Clawdi dashboard — it's the first thing offered after login it appears under Agents; find its environment id with clawdi project list --include-envs (an environment id is a project id)
Hermes (local researcher) Hermes Agent: curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash, then hermes setup hermes status shows a model provider with usage available (the most common blocker in practice)
An orchestrator Claude Code or Codex — interchangeable throughout this repo opened at the repo root, where both auto-read AGENTS.md
The skills npx skills add FIL-Builders/filecoin-clawdi-fleet -g npx skills ls -g lists fleet-setup

Cloning is optional and only for working on the repo — the skills are self-contained, so every install route below works without one: git clone https://github.com/FIL-Builders/filecoin-clawdi-fleet.git.

2. Automated setup — one instruction to your orchestrator

Open Claude Code or Codex at the repo root and say:

Set up the fleet.

That triggers the fleet-setup skill, which detects what's installed and wires every layer, verifying as it goes:

  1. Register — clawdi setup --no-daemon (per-agent MCP wiring), confirmed against clawdi doctor's Environments line. --no-daemon matters: the background daemon would mirror every session of every registered agent to the cloud, and the pipeline shares state through memory receipts and Filecoin blobs instead. If one agent fails with API error 500 while the others register, the targeted form works where the sweep doesn't: clawdi setup --no-daemon --agent <type>.
  2. Publish the skills — clawdi skill add skills/<name> -p personal -y for fleet-creator, fleet-setup, memory-researcher, memory-consolidator, then clawdi pull --modules skills -p personal --agent <type> per local agent (idempotent, content-hashed — re-running is safe).
  3. Wire the storage rail — foc-cli wallet init --keyRef clawdi:FILECOIN_PRIVATE_KEY, then wallet fund and wallet deposit 1. foc-cli stores only the reference; the key stays in the vault and is resolved per command. The faucet allows roughly one claim per address per day; if it refuses, fund the address from wallet balance --json in a browser and re-run the deposit.
  4. Prove it — the four-step ladder in storage-vault.md, including the negative test, then a store/recall smoke test, then the cross-agent check: a different agent finds the FILECOIN-MEMORY smoke-test receipt.

The vault key itself is a human step, done once: clawdi vault set FILECOIN_PRIVATE_KEY --prompt — never an agent, never from a chat. Use a dedicated low-value wallet.

Session privacy. With --no-daemon, nothing uploads automatically. To publish the pipeline's sessions deliberately: clawdi push --modules sessions --agent claude_code from the repo root — project-scoped, only this repo's sessions. Hermes ignores project filters (its push is all-or-nothing) — leave its sessions local.

Prefer doing it by hand? Every command lives in the skill's references, one file per agent and layer.

3. The hosted OpenClaw box — one attach, then four commands in its terminal

First, from your machine (or ask your orchestrator), attach the vault to the box's project — the environment id from the prerequisites table:

clawdi vault attach default --project <hosted-environment-id>

Then open the hosted agent in the Clawdi dashboard and run in its terminal — no repo clone needed:

clawdi pull --modules skills -p personal --agent openclaw
npm install -g foc-cli
foc-cli wallet init --keyRef clawdi:FILECOIN_PRIVATE_KEY   # the box's OWN reference
foc-cli wallet balance --json                              # keySource "keyRef" + address

(If the box shows unauthenticated, fix the credential from the dashboard — never by pasting tokens. References resolve per project, so derive them in the box; never copy a config in from another machine.)

4. Done when

clawdi doctor                   # every intended agent on the Environments line
foc-cli wallet balance --json   # keySource "keyRef" on your machine AND in the box

Something failing? troubleshooting.md.

5. Next step — your first pipeline

The fleet-setup skill always ends its report the same way: a checklist of anything that still needs you (hosted-box terminal steps, manual faucet funding, a decision about any pre-existing wallet config), then a prompt to come back once those are done. When they are, say to your orchestrator:

Set up a pipeline.

That starts pipelines.md. You will not write a config file: the fleet-creator skill asks what you want watched and why, offers worked examples to react to, and turns your answers into a spec. It runs one supervised cycle first — you watch each agent's tick and the stored blob verify — then creates the crons that keep it running on its own.