Forge — Build Production-Ready OpenClaw Agents and Skills

Every OpenClaw agent starts the same way: you create a workspace, write SOUL.md, set up AGENTS.md, wire memory files, configure tools... and somewhere around file #8 you realize the structure is wrong or you forgot tools.deny and your agent can restart the gateway and kill itself.

Forge is a senior agent architect who's seen every mistake. It guides you through creating production-ready agents and skills via conversation — not templates you'll abandon, but working systems with memory, self-improvement, and anti-patterns baked in from day one.

What you get

• Guided skill creation — 11-step workflow from idea to working SKILL.md with proper frontmatter, trigger phrases, and examples
• Full agent scaffolding — 9-step process that generates all 13 workspace files with real content, not empty templates
• 22 pitfall warnings — catches silent failures like YAML without fences, model aliases, data in memory/, missing tools.deny
• 4-level memory architecture — contextual → file-based → vector → identity, wired correctly from the start
• Self-improvement engine — patterns.md → lessons.md promotion loop so your agents actually learn from mistakes
• Audit mode — review any existing agent or skill for quality issues and structural problems
• Type-first design — Workflow, Role, Data-driven, or Hybrid for skills; Full, Specialized, or Mask for agents

Setup in 3 minutes

1. openclaw agents add forge — install the agent
2. Copy workspace/ to your OpenClaw agents directory
3. Start a new chat with /new or send a message
4. Forge runs onboarding (name, timezone, experience level)
5. Use /new-skill or /new-agent to start building

Built for

• Developers building their first OpenClaw agent who don't want to learn the hard way
• Teams scaling from 1 agent to 5+ who need consistent architecture across bots
• Anyone tired of debugging silent config failures and empty template files

Production agents shouldn't be a guessing game. Forge makes sure they aren't.