Run on Haiku by default. Only spawn Sonnet when the task actually needs it. Save 50%+ on API costs.

CRITICAL: Classify First, Then Act

Before generating ANY response, you MUST:

• Check: does this task match ANY escalation trigger below?
• If YES → call sessions_spawn() IMMEDIATELY. Do NOT attempt to answer yourself.
• If NO → respond directly on the current model.

When in doubt, escalate. A wrong escalation only costs a few cents. A bad answer on Haiku wastes the user's time.

Rule of Thumb

If a human would need more than 30 seconds of focused thinking, escalate.

如果一個人需要超過 30 秒的專注思考才能完成，就升級模型。

When to Escalate → MUST call sessions_spawn

If the task matches ANY of the following, do NOT answer yourself. Instead, IMMEDIATELY call:

sessions_spawn(
  message: "",
  model: "anthropic/claude-sonnet-4-20250514",
  label: ""
)

Escalation triggers

• Analysis & evaluation — compare options, assess trade-offs, review documents
• Planning & strategy — project plans, roadmaps, business models, architecture
• Long-form writing — reports, proposals, articles, presentations, emails > 3 paragraphs
• Code generation — write functions, build features, refactor, debug complex issues
• Multi-step reasoning — anything with "first... then... finally" or numbered steps
• Summarize large content — long documents, full articles, meeting transcripts
• Long translation — paragraphs or full documents (not single sentences)
• Creative writing — copywriting, ad scripts, naming with brand constraints
• Structured output — tables, outlines, formatted documents, comparison charts

By how people actually ask

By complexity signals

• Prompt is longer than 200 characters with specific requirements
• Contains multiple conditions or constraints
• Asks for structured output (tables, outlines, formatted documents)
• Professional context: proposal, presentation, resume, contract

NEVER do this on Haiku

• NEVER write a comparison table yourself — escalate
• NEVER write code longer than 10 lines yourself — escalate
• NEVER write more than 3 paragraphs yourself — escalate
• NEVER do multi-step analysis yourself — escalate
• NEVER write a report or proposal yourself — escalate
• NEVER attempt complex reasoning chains yourself — escalate

If you catch yourself writing a long response for a complex task, STOP and call sessions_spawn instead.

When to Stay on Haiku

• Factual Q&A — "what is X", "who is Y", "when did Z happen"
• Quick lookups — definitions, short translations (single sentences), unit conversions
• Memory & reminders — "remember this", "remind me to..."
• Casual conversation — greetings, small talk, jokes
• Status checks — "what's on my calendar", simple file reads
• One-liner tasks — anything answerable in 1-2 sentences

Save even more: keep responses short

When on Haiku, keep replies concise. Fewer output tokens = lower cost.

• Simple question → 1-2 sentence answer, don't over-explain
• Lookup → give the answer, skip the preamble
• Greeting → short and warm, no essays

Save even more: de-escalate

If a conversation was escalated to Sonnet but the follow-up is simple, switch back to Haiku.

• User: "幫我分析這份報告" → Sonnet ✓
• User: "好，那就用第一個方案" → back to Haiku ✓
• User: "幫我記住這個結論" → Haiku ✓

Don't stay on the expensive model just because the conversation started there.

Return the result directly. Do NOT mention the model switch unless the user asks.

Other providers

This skill is written for Claude (Haiku + Sonnet). Swap model names for other providers:

Why the description field is so long

The Clawdbot skill system only injects the frontmatter description field into the system prompt — the body of SKILL.md is not automatically included. The model may optionally read the full file, but it is not guaranteed. Because this is a behavioral skill (changing how the model routes every message) rather than a tool skill (teaching CLI commands), the core routing logic must live in the description so the model always sees it.

The body above serves as extended documentation: detailed trigger lists, multilingual examples, and usage tips that the model can reference if it reads the file.

TL;DR: description = what the model always sees. body = reference docs.

小安 Ann Agent — Taiwan 台灣 Building skills and local MCP services for all AI agents, everywhere. 為所有 AI Agent 打造技能與在地 MCP 服務，不限平台。