Agent Builder

**Quick judgment** A strong, opinionated builder for OpenClaw agents that closely follows the canonical workspace structure. It is especially useful for users who already understand OpenClaw concepts and want to move fast while preserving safety and consistency. **Strengths** - Produces a full minimal-viable OpenClaw workspace (IDENTITY, SOUL, AGENTS, USER, HEARTBEAT, optional MEMORY + dated memory files, optional TOOLS) using templates tied to official references. - Uses a focused, multi-round interview to define mission, surfaces, autonomy level, prohibitions, memory policy, tone, and tool posture. - Explicitly enforces a guardrail checklist (ask-before-destructive, ask-before-outbound messages, CLI error-recovery guidance, loop-breaking, group-chat etiquette). - Provides concrete acceptance-test prompts so users can quickly validate the resulting agent’s behavior. **Key risks / limitations** - Assumes familiarity with OpenClaw concepts; novices may under-specify critical constraints or misinterpret autonomy levels, leading to overly permissive agents. - Templates and “strong defaults” may diverge from evolving OpenClaw best practices over time if not actively maintained. - Safety and behavior quality still depend heavily on the user’s answers; vague or conflicting requirements can produce inconsistent or leaky guardrails. - Advanced or highly specialized domains (complex toolchains, strict compliance regimes) may require substantial manual refinement beyond what the generic templates capture. **Recommended scenarios** - You want to spin up a new OpenClaw agent quickly with clear safety rules, autonomy boundaries, and heartbeat behavior. - You have an existing agent that loops, overreaches, or behaves poorly in groups and want targeted, minimal diffs to SOUL/AGENTS/HEARTBEAT. - A team needs a repeatable, documented pattern for agents so future workspaces are consistent and auditable. - You want ready-made acceptance tests to regression-test an agent after changes to its rules or persona.