# Official Claude Fable 5 — 12 Prompting Patterns

Source: https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-fable-5
Retrieved: June 2026 (model suspended June 12 — docs still live)

## The 12 Official Patterns

| # | Pattern | Purpose | Verbatim Instruction |
|---|---------|---------|---------------------|
| 1 | **Longer turns by default** | Prevent overplanning on long-running tasks | "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** | Prevent unrequested refactoring at high effort | "Don't add features, refactor, or introduce abstractions beyond what the task requires. A bug fix doesn't need surrounding cleanup." |
| 3 | **Strong instruction following** | Brevity via intent, not enumeration | "Lead with the outcome. Your first sentence should answer 'what happened' or 'what did you find'." |
| 4 | **Ground progress claims** | Eliminate 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." |
| 5 | **State the boundaries** | Prevent unrequested actions | "When the user is describing a problem, asking a question, or thinking out loud rather than requesting a change, the deliverable is your assessment." |
| 6 | **Parallel subagents** | Delegate independent work | "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 lesson storage | "Store one lesson per file with a one-line summary at the top. Record corrections and confirmed approaches alike." |
| 8 | **Rare cases of early stopping** | Autonomous mode reminder | "You are operating autonomously. The user is not watching in real time and cannot answer questions mid-task. For reversible actions, proceed without asking." |
| 9 | **Rare cases of context-budget concern** | Prevent premature summarization | "You have ample context remaining. Do not stop, summarize, or suggest a new session on account of context limits." |
| 10 | **Give the reason, not only the request** | Provide intent context | "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 when communicating with user** | Distinguish working notes from final output | "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." |
| 12 | **Create a send-to-user tool** | Verbatim mid-task message delivery | Custom tool with `{"name": "send_to_user", "input_schema": {"message": "string"}}`. Must be paired with an instruction to actually call it. |

## Patterns NOT in the Fable Brain Kit

The Hyperautomation Labs "Fable Brain Kit" (PDF) condensed patterns 1-5 into 8 rules but **omitted patterns 6-12 entirely**:

| Missing from Kit | Official Pattern # | Why It Matters |
|---|---|---|
| Parallel subagents delegation | 6 | Without explicit instruction, models don't delegate automatically even when capable |
| File-based memory system | 7 | Single-file memory blobs are less effective than per-lesson files with one-line summaries |
| Autonomous mode reminder | 8 | Prevents mid-task blocking for permission on reversible actions |
| Context-budget reassurance | 9 | Prevents premature session abandonment in long contexts |
| Give the reason | 10 | User-side technique, not agent-side behavior |
| Readability for user | 11 | Eliminates arrow-chain shorthand and jargon in final messages |
| send_to_user tool | 12 | Plugin-level tool, not applicable to standard Hermes sessions |

## Recommended Scaffolding Changes

From the official guide's closing section:

1. **Start at the top of your difficulty range** — Pick a task harder than what you'd assign to prior models
2. **Make self-verification explicit** — Separate, fresh-context verifier subagents outperform self-critique
3. **Refactor existing prompts and skills** — Prior-model prompts are often too prescriptive for stronger models
4. **Don't instruct reasoning reproduction** — Echoing/transcribing internal reasoning can trigger refusal classifiers
5. **Create a send-to-user tool** — For long async agents needing verbatim mid-task delivery