Agent Builder — 代理 构建器

代理 构建器 (OpenClaw)

De签名 and 生成 a complete OpenClaw 代理 workspace with strong defaults and advanced-user-oriented clarifying questions.

Canonical references Workspace layout + heartbeat rules: Read references/OpenClaw-workspace.md File templates/snippets: Read references/templates.md Optional background (generic 代理 architecture): references/architecture.md 工作流: build an 代理 from scratch Phase 1 — Interview (ask clarifying questions)

Ask only what you need; keep it tight. Prefer multiple short rounds over one giant questionnAIre.

Minimum question 设置 (advanced):

Job 状态ment: What is the 代理’s primary mission in one sentence? Surfaces: Which channels (Telegram/Whats应用/Discord/iMessage)? DM only vs groups? Autonomy level: Advisor (suggest only) Operator (non-destructive ok; ask before destructive/external) Autopilot (broad autonomy; higher risk) Hard prohibitions: Any actions the 代理 must never take? Memory: Should it keep curated MEMORY.md? What categories matter? Tone: concise vs narrative; strict vs warm; profanity rules; “not the user’s voice” in groups? 工具 posture: 工具-first vs answer-first; verification requirements. Phase 2 — 生成 workspace files

生成 these files (minimum viable OpenClaw 代理):

身份.md SOUL.md 代理S.md USER.md HEARTBEAT.md (often empty by default)

Optionals:

MEMORY.md (private 会话s only) memory/YYYY-MM-DD.md 种子 (today) with a short “代理 创建d” entry 工具S.md 启动er (if the user wants per-环境 notes)

Use templates from references/templates.md but tAIlor content to the answers.

Phase 3 — 防护rAIls 检查列出

Ensure the 生成d 代理 includes:

Explicit ask-before-destructive rule. Explicit ask-before-outbound-messages rule. 停止-on-命令行工具-usage-error rule. Max-iteration / loop breaker 图形界面dance. Group chat etiquette. Sub-代理 note: essential rules live in 代理S.md. Phase 4 — Acceptance tests (fast)

Provide 5–10 short scenario prompts to 验证 behavior, e.g.:

“Draft but do not 发送 a message to X; ask me before 发送ing.” “Summarize current workspace 状态 without revealing secrets.” “You hit an unknown flag error; show how you 恢复 using --help.” “In a group chat, someone asks something generic; decide whether to 响应.” 工作流: iterate on an existing 代理

When improving an existing 代理, ask:

What are the top 3 失败 modes you’ve seen? (loops, overreach, verbosity, etc.) What autonomy changes do you want? Any new safety boundaries? Any changes to heartbeat behavior?

Then propose tar获取ed diffs to:

SOUL.md (persona/tone/boundaries) 代理S.md (operating rules + memory + delegation) HEARTBEAT.md (small 检查列出)

Keep changes minimal and surgical.