PicoClaw Fleet

Orchestrate a fleet of remote PicoClaw workers over SSH for fast, ephemeral one-shot tasks.

Purpose

Use this skill to deploy PicoClaw to remote machines, dispatch one-shot tasks, fan out work in parallel, and optionally tear workers down after completion.

Skill Files

• scripts/deploy.sh — install/update PicoClaw on a host
• scripts/dispatch.sh — run picoclaw agent -m "TASK" on a host and return stdout
• scripts/fleet-status.sh — check host reachability and install readiness

1) Always read fleet config first

Fleet config path:

• ~/.openclaw/workspace/config/picoclaw-fleet.json

If missing, create it with this default template:

{
  "hosts": [
    {
      "name": "darth",
      "host": "192.168.50.57",
      "user": "eric",
      "arch": "arm64",
      "ssh_key": "~/.ssh/id_rsa"
    }
  ],
  "defaults": {
    "provider": "anthropic",
    "api_key_env": "ANTHROPIC_API_KEY"
  }
}

2) Deploy PicoClaw to a host

Use scripts/deploy.sh <host> <user> <arch> [ssh_key].

Expected behavior:

• Resolve latest release from EricGrill/picoclaw GitHub releases
• Select architecture asset (amd64, arm64, riscv64)
• Install binary remotely to ~/.local/bin/picoclaw
• Create ~/.picoclaw/.env with provider + API key env value
• Run picoclaw onboard

Required envs before deploy:

• ANTHROPIC_API_KEY (default) or whichever env is set by defaults.api_key_env
• Optional: PROVIDER (defaults to anthropic)
• Optional: API_KEY_ENV override

3) Dispatch one-shot work

Use scripts/dispatch.sh <host> <user> <task> [timeout_seconds].

Behavior:

• SSH into host
• Run picoclaw agent -m "TASK"
• Enforce timeout (default 120s)
• Return stdout directly (clean output for inline display)

4) Run tasks in parallel across hosts

For multi-task batches, dispatch to multiple hosts in background and wait for all:

scripts/dispatch.sh 192.168.50.57 eric "summarize logs" 120 > /tmp/darth.out 2>&1 &
scripts/dispatch.sh 192.168.50.58 eric "extract action items" 120 > /tmp/lobot.out 2>&1 &
wait

Then print each host output inline.

5) Host selection policy

For single tasks:

1. Run scripts/fleet-status.sh
2. Prefer reachable hosts where picoclaw is installed
3. Pick least-loaded host when load data exists; otherwise pick first available

If selected host is missing PicoClaw, run deploy first.

6) Teardown (optional)

To remove PicoClaw after one-shot jobs:

ssh -i ~/.ssh/id_rsa eric@HOST 'rm -f ~/.local/bin/picoclaw ~/.picoclaw/.env'

Use teardown only when explicitly requested or for strict ephemeral execution workflows.

Failure handling

• SSH failure: report host as unreachable and continue with other hosts
• Deploy failure on one host: continue dispatching to healthy hosts
• Timeout: return timeout status with partial output if present
• Missing config: create template, then re-run

Recommended workflow

1. Load/validate fleet config
2. Check health: scripts/fleet-status.sh
3. Deploy missing hosts: scripts/deploy.sh ...
4. Dispatch task(s): scripts/dispatch.sh ...
5. Aggregate outputs and return inline
6. Optional teardown if requested