# Full Official Fable 5 Prompting Patterns

Researched from Anthropic's official guide "Prompting Claude Fable 5" at `platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-fable-5`. Retrieved 2026-06-15.

## The 12 Official Patterns

### 1. Longer Turns by Default
Hard tasks run many minutes at higher effort. Adjust timeouts, streaming, and progress indicators before migrating. Use async harnesses.

> "When you have enough information to act, act. Do not re-derive facts already established in the conversation, re-litigate a decision the user has already made, or narrate options you will not pursue."

### 2. Consider All Effort Levels
Use `high` as default. `xhigh` for most capability-sensitive work. `medium`/`low` for routine.

To prevent unrequested refactoring at higher effort:
> "Don't add features, refactor, or introduce abstractions beyond what the task requires. A bug fix doesn't need surrounding cleanup. Don't design for hypothetical future requirements."

### 3. Strong Instruction Following
Brief instructions suffice — no need to enumerate every case.

Brevity instruction:
> "Lead with the outcome. Your first sentence after finishing should answer 'what happened' or 'what did you find': the thing the user would ask for if they said 'just give me the TLDR.'"

Checkpoint instruction:
> "Pause for the user only when the work genuinely requires them: a destructive or irreversible action, a real scope change, or input that only they can provide."

### 4. Ground Progress Claims During Long Runs
Nearly eliminated fabricated status reports:
> "Before reporting progress, audit each claim against a tool result from this session. Only report work you can point to evidence for; if something is not yet verified, say so explicitly."

### 5. State the Boundaries
> "When the user is describing a problem, asking a question, or thinking out loud rather than requesting a change, the deliverable is your assessment. Report your findings and stop."

### 6. Parallel Subagents
Claude Fable 5 dispatches parallel subagents more readily. Use subagents frequently:
> "Delegate independent subtasks to subagents and keep working while they run. Intervene if a subagent goes off track."

### 7. Construct a Memory System
File-based memory significantly boosts performance:
> "Store one lesson per file with a one-line summary at the top. Record corrections and confirmed approaches alike."

To bootstrap:
> "Reflect on the previous sessions we've had together. Use subagents to identify core themes and lessons."

### 8. Early Stopping (Rare)
For autonomous pipelines:
> "You are operating autonomously. The user is not watching in real time and cannot answer questions mid-task. For reversible actions that follow from the original request, proceed without asking."

### 9. Context-Budget Concern (Rare)
> "You have ample context remaining. Do not stop, summarize, or suggest a new session on account of context limits."

### 10. Give the Reason
> "I'm working on [the larger task] for [who it's for]. They need [what the output enables]. With that in mind: [request]."

### 11. Readability for User
> "Terse shorthand is fine between tool calls. Your final summary is different: it's for a reader who didn't see any of that. Write it as a re-grounding: the outcome first, then the one or two things you need from them."

> "When you write the summary at the end, drop the working shorthand. Write complete sentences. Spell out terms. Don't use arrow chains. Open with the outcome."

### 12. Create a Send-to-User Tool
A custom tool for verbatim mid-task delivery:
```json
{
  "name": "send_to_user",
  "description": "Display a message directly to the user. Use for progress updates, partial results, or content the user must see exactly as written.",
  "input_schema": {
    "type": "object",
    "properties": {
      "message": { "type": "string" }
    },
    "required": ["message"]
  }
}
```
Pair with:
> "Between tool calls, when you have content the user must read verbatim, call the send_to_user tool with that content."

## Scaffolding Changes Recommended

1. **Start at top of difficulty range** — pick tasks harder than what prior models could handle
2. **Make self-verification explicit** — separate verifier subagents outperform self-critique
3. **Refactor existing prompts** — prior-model skills are too prescriptive for Fable-5-class models
4. **Don't instruct reasoning reproduction** — triggers refusal classifiers; use structured thinking blocks
5. **Create a send-to-user tool** — for long asynchronous agents

## Availability Note

Claude Fable 5 was suspended June 12, 2026 (US government export control). The prompting patterns apply to any agent-capable model, including DeepSeek V4 Flash, Qwen 3.6, and Gemma 4.