Coder for OpenClaw

Install and wire a coding-focused OpenClaw sub-agent for background code execution, test-driven edits, bug fixing, small project scaffolding, and small-to-medium data-analysis tasks.

What this skill is

This is an integration skill for installing and wiring the coder OpenClaw sub-agent from the public repository:

• coder-openclaw-agent

The repository contains:

• the workspace-coder prompt pack;
• the coder-sandbox:latest Docker image definition;
• the coder agent config template;
• the Main -> Coder orchestration contract.

This skill is intended for OpenClaw users who want a strong background coding and data-analysis sub-agent without building the orchestration from scratch.

What it can do

• code execution and verification inside the OpenClaw sandbox;
• bug fixing and test-driven edits;
• small project scaffolding;
• small-to-medium data-analysis tasks;
• HTML, PDF, spreadsheet, and office-style document processing;
• honest blocked-state reporting through PARTIAL or FAILURE.

Requirements

• OpenClaw 2026.3.x or later
• Docker available on the host
• an authenticated openai-codex provider profile

Install

1. Clone the repository:
   • git clone https://github.com/MilleniumGenAI/coder-openclaw-agent.git
2. Copy openclaw/workspace-coder/ into your OpenClaw base directory, or point your agent config at that path directly.
3. Build the sandbox image from the repository root:
   • docker build -f docker/coder-sandbox.dockerfile -t coder-sandbox:latest .
4. Register the agent in openclaw.json using:
   • openclaw/agent-config.template.json
5. If your main agent delegates coding tasks, align it with:
   • openclaw/main-coder-prompt.md

Validate

Run these checks before using the agent in real work:

bash

openclaw models status --agent coder --probe --probe-provider openai-codex --json
openclaw sandbox explain --agent coder

Then run a first smoke task:

bash

openclaw agent --agent coder --json --message "Return strictly valid JSON matching coder SOUL schema. GOAL: create /tmp/coder/smoke/main.py that prints hello. INPUTS: none. CONSTRAINTS: work only in /tmp/coder/smoke; use python3 and Linux/bash commands only; use PARTIAL if blocked. SUCCESS CRITERIA: python3 /tmp/coder/smoke/main.py prints hello. DELIVERABLES: codeblocks and sandbox_log."

Core references

• Root README: README.md
• Agent config template: openclaw/agent-config.template.json
• Main -> Coder orchestration guide: openclaw/main-coder-prompt.md
• Runtime inventory: docker/RUNTIME.md
• Known limits: docs/known-limits.md

Notes

• This is an OpenClaw-only v1 package.
• ClawHub publishes skills under platform-wide MIT-0 terms.
• The runtime source of truth is openclaw/workspace-coder/SOUL.md.
• Default working area inside the sandbox is /tmp/coder/<task_name>/.
• The expected output contract is strict JSON with SUCCESS | PARTIAL | FAILURE.