From openclaw
Full guided journey to create a complete OpenClaw agent workspace from scratch. Chains discovery, memory design, persona crafting, workspace scaffolding, and quality audit into a single orchestrated flow with checkpoints. Use when the user says "create my agent from scratch", "full agent setup", "build a complete agent", "new openclaw agent", "workspace wizard", "guided agent creation", or "walk me through creating an agent". This is the comprehensive path — for quick single-file edits use openclaw-agent-architect instead.
How this skill is triggered — by the user, by Claude, or both
Slash command
/openclaw:openclaw-workspace-wizardThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
The full guided journey for creating a production-ready OpenClaw agent workspace. Chains six phases with user checkpoints between each, so you can pause, adjust, and resume at any point.
The full guided journey for creating a production-ready OpenClaw agent workspace. Chains six phases with user checkpoints between each, so you can pause, adjust, and resume at any point.
This wizard orchestrates the complete lifecycle:
Phase 1: Discovery → What agent do you need?
Phase 2: Memory Design → How should it remember?
Phase 3: Soul Design → Who is this agent?
Phase 4: Workspace → What files and structure?
Phase 5: Scaffolding → Generate everything
Phase 6: Quality Audit → Score and improve
Each phase produces artifacts that feed the next. You can skip phases if you already have the artifacts (e.g., skip Discovery if you already know exactly what you want).
Determine workspace settings:
~/.openclaw/workspace/ (single-agent)~/.openclaw/agents-workspaces/<agent-id>/Check the workspace location for existing artifacts:
"Welcome! Let's build your OpenClaw agent workspace step by step.
Your settings:
This journey has 6 phases. I'll check in with you between each one so you can adjust course. Ready?"
Wait for user confirmation.
Goal: Understand what the agent does, who it serves, and what it needs.
"Before we dive in — would you like to browse reference templates for common agent archetypes? I have templates for Software Architect, Code Reviewer, Security Engineer, Technical Writer, DevOps Automator, Sprint Prioritizer, UX Researcher, and Feedback Synthesizer.
[RT] Browse templates for inspiration, or just tell me about your agent."
If RT selected: Read and present references/agent-templates/index.md, let user browse individual templates, then return here.
Ask the user:
Probe deeper based on responses:
Walk through 3-5 concrete scenarios:
Based on the conversation, determine:
"Phase 1 complete! Here's what I've captured:
[C] Continue to Phase 2 (Memory Design) [R] Revise something [M] Save progress and stop here"
Goal: Design the memory architecture before building the persona.
Based on the discovery:
Discuss trade-offs and confirm the strategy.
What categories should MEMORY.md have?
If daily logs enabled:
"Phase 2 complete! Memory architecture:
[C] Continue to Phase 3 (Soul Design) [R] Revise something [M] Save progress and stop here"
Goal: Craft a coherent, opinionated, token-efficient SOUL.md.
Craft collaboratively: "You are [NAME], a [ROLE] for [CONTEXT]."
One sentence. Specific. Predictive of behavior.
Each value needs:
Anti-patterns to avoid: "Be helpful" (too vague), "Ensure quality" (not actionable).
Probe for:
Each boundary must be specific and actionable.
Design the communication personality:
SOUL.md must be under 2,000 words. Present the complete draft and word count. If over, suggest specific cuts.
Present the complete SOUL.md draft for review.
"Phase 3 complete! Your SOUL.md is {word_count} words.
[C] Continue to Phase 4 (Workspace Design) [R] Revise something [M] Save SOUL.md and stop here"
Goal: Design IDENTITY.md and determine the complete file manifest.
Create collaboratively:
Design the operating manual:
For each file identified in Phase 1:
"Phase 4 complete! Workspace manifest: {list all files to generate} {directory structure diagram}
[C] Continue to Phase 5 (Scaffolding) [R] Revise something [M] Save progress and stop here"
Goal: Create all workspace files and validate them.
For single-agent: flat structure at workspace path. For multi-agent: per-agent directories with shared files.
Write each file using the design decisions from Phases 1-4:
Run checks:
Report any issues.
"Phase 5 complete! Generated {N} files at {workspace_path}: {file list with sizes}
[C] Continue to Phase 6 (Quality Audit) [R] Fix something before audit [M] Stop here — workspace is ready for use"
Goal: Run a comprehensive quality audit on the generated workspace.
Invoke the openclaw-workspace-audit skill logic on the workspace we just created. This provides an independent quality check of everything we built.
Present the scored report and fix any issues identified.
"Your OpenClaw workspace is ready!
Workspace: {path} Quality Score: {score}/200 — {rating} Files: {list}
Next steps:
/oc-workspace-audit to reassess quality/oc-autonomy if you want to enable autonomous operation/oc-memory if you want to fine-tune memory architectureQuick commands:
/oc-architect — Modify individual workspace files/oc-workspace-audit — Re-run quality audit anytime/oc-autonomy — Autonomy readiness assessment"This wizard uses content and patterns from:
references/agent-templates/)npx claudepluginhub theophiluschinomona/claude-openclaw-plugin --plugin openclawGuides creation of Claude Code agents via 5 phases: analyze requirements and type, design architecture with patterns/skills, create files/frontmatter, validate, refine iteratively.
Asks 10 discovery questions to understand user needs, configures AI agent workspace files for target system, tests integrations, and implements security guardrails. Use for new agent setups.
Create custom agents for Claude Code including YAML frontmatter, system prompts, tool restrictions, and discovery optimization. Use when creating, building, or designing agents, or when asked about agent creation, subagent configuration, Task tool delegation, or agent best practices.