Where this came from: Modules 2.1 to 2.6 came from the user’s own practical guide on how to use Fable 5 as a work agent, not from an official Anthropic citation—each prompt there is tested in practice, not “verbatim from the docs.” This bonus module wraps up Track 2 and the entire course: first, a dense recap of the 12 tips, then the master prompt that combines them all, and finally a map showing how this practical track connects to the technical/official foundation of track 1.
Recap: the 12 tips from track 2
Imagine it this way: track 1 gave you the grammar of how to talk to Fable 5. Track 2 gave you the manual for how to work with it—choose the right task, provide context, control depth, verify for real, delegate by role, preserve memory. Together, the two become a single way of working.
Before the master prompt, it’s worth consolidating the 12 tips from modules 2.1 to 2.6 into one list—so you can scan in seconds to see which one is missing from your current session. Each line includes its source module in brackets, in case you want to go back and reread the full context.
The grammar of track 1 (what to say) plus the work manual of track 2 (how to operate) form the same work system that the master prompt encodes in topic 2.
CHEAT SHEET — Fable 5 como Agente de Trabalho (módulos 2.1–2.6) [2.1] Use o Fable 5 pra problemas difíceis — arquitetura, risco, plano, auditoria — não pra resumir texto [2.1] Explique o motivo do pedido — objetivo, público e resultado esperado, antes da tarefa [2.2] Controle a profundidade — baixa/média/alta/máxima conforme a complexidade real [2.2] Defina limites antes de deixar o agente agir — diga o que ele pode e não pode fazer [2.3] Exija verificação real — evidência de ferramenta, nunca "pronto" sem prova [2.3] Use subagentes por especialidade — estratégia, produto, comunicação, tecnologia, cético [2.4] Crie memória de projeto — um arquivo por tema, atualize em vez de duplicar [2.4] Mande o agente continuar quando já houver informação suficiente — não termine com promessa [2.5] Use visão pra revisar interfaces, prints e criativos — não só "deixe mais bonito" [2.5] Peça revisão defensiva pra segurança — nunca exploração/ataque [2.6] Crie loops com verificação — foco, critério, teste, aprendizado, não só repetição [2.6] Use memória externa pra trabalhos longos — não dependa só do histórico do chat
In one sentence: 12 tips, one list — if your session doesn’t include at least half of them, it’s still in chat mode.
The complete master prompt
This isn’t a prompt about a task. It’s a prompt about how Fable 5 should operate for any task of yours—the synthesis of the 12 tips in a single generic block, tested in practice. Paste it at the start of a new session (or pin it in a CLAUDE.md/system prompt) and it becomes that project’s “constitution.”
The master prompt doesn’t ask for an answer: it sets up five lenses that converge on a single recommendation, always with a concrete next step.
Goal: turn Fable 5 into a permanent strategic and operational agent for your project—one that diagnoses, decides, executes, verifies, and improves instead of just responding to one-off requests.
You are my strategic and operational agent. Context: I'm working on a project that needs to turn ideas into practical execution, with clarity, quality, and real results. Objective: Create a work system that diagnoses problems, sets priorities, executes tasks, verifies results, and improves continuously. How to work: 1. Don't give me loose ideas. 2. Think like a business, product, content, technology, and execution architect. 3. Use subagents: strategy, product, communication, technology, and skeptic. 4. Always deliver a diagnosis, recommended decision, execution plan, and first next step. 5. If you already have enough information, act. Don't end with a promise. 6. When making claims, say what is evidence, what is a hypothesis, and what needs to be tested. 7. Record important decisions in project memory. Task: Analyze my current project and propose the simplest architecture for turning intent into execution, ongoing practice, and measurable results.
How to check: paste the entire block (without editing it the first time, just to see the default behavior) and check whether the response includes the 4 parts promised in item 4—diagnosis, recommended decision, plan, and first next step—instead of a loose paragraph of ideas. If anything is missing, reread item 4 of “Working mode”: that's what enforces the structure.
⚠️ Attention
Don’t run this prompt without editing the block Context e Goal for your actual project—without that, the answer stays generic because the prompt itself is generic. Topic 3 shows exactly what to change.
💡 Practical tip
Test it first on a small, familiar project where you already know what the right answer would be. That gives you a quick benchmark to see whether the master prompt is calibrated before you trust it with a large, unfamiliar project.
In one sentence: The master prompt doesn’t ask “what do you think?” It builds a framework for diagnosis, decision, plan, and action—and makes the model fill it out every time.
How to customize it for your project
The master prompt has two parts of different kinds: a variable core (context, objective, subagent roles) that changes from project to project, and a fixed structure (diagnosis → decision → plan → step), which is exactly what makes it work as a work system, not as a chat. Replacing the first part is mandatory; changing the second usually makes the result worse.
Read the entire master prompt once
Before editing any line, understand the 7 parts of “Work Mode”—this is the list that carries the discipline of the 12 tips.
Replace the Context block
Describe your actual project, your audience, and what the result needs to enable — just like tip 2 (explain the reason) in module 2.1.
Replace the Objective block
In one sentence, say what you need the work system to do for you — not today’s task, but the agent’s permanent role in the project.
Adjust the subagent roles
"Strategy, product, communications, technology, skeptic" is intentionally generic — replace these with roles that actually exist in your domain (e.g., legal, UX, data, operations).
Keep the structure intact
Diagnosis → decision → plan → next step, plus items 5, 6, and 7 (act with enough information, separate evidence from hypotheses, record in memory) stay the same — they’re what make the prompt work.
Contexto: Estou trabalhando em <nome do projeto/produto> para <seu público-alvo>. O resultado final precisa permitir <o objetivo real que o público vai alcançar>. Objetivo: Criar <o que o sistema de trabalho deve fazer por você, em uma frase>. 3. Use subagentes: <papel 1>, <papel 2>, <papel 3>, <papel 4> e cético.
✗ Common customization mistake
- ✗Replace only the Objective and leave the Context generic—the model still won’t know who it’s for.
- ✗Delete item 5 (“act if you already have enough information”), thinking it’s “too much autonomy.”
- ✗Keeping the 5 generic subagent roles in a technical project where they don’t make sense.
✓ The right approach
- ✓Context and Objective always together, both specific to your project.
- ✓Keep the 7-item “Working mode” structure intact—it’s what keeps the response focused.
- ✓Subagent roles named using the actual language of your domain.
In one sentence: change the context, objective, and subagent roles; keep the diagnosis → decision → plan → step structure intact.
Chat vs. work system
The entire learning path 2—and this module in particular—exists to highlight a difference: using Fable 5 by asking it to responses versus using Fable 5 as system. The difference isn't in the model's intelligence. It's in how much context, memory, verification, and continuous execution you build around it.
✗ Chat mode
- ✗Every question starts from scratch, with no accumulated context.
- ✗Just one opinion, without verification or a counterpoint.
- ✗"Done" is what the model says, not what has been proven.
- ✗Nothing gets recorded—the next session repeats the same mistake.
✓ Work system
- ✓Explicit context (the why) before each task.
- ✓Subagents by role, consolidated into a single recommendation.
- ✓"Done" requires real evidence from a tool.
- ✓Project memory — today’s lesson becomes tomorrow’s starting point.
The value of Fable 5 isn’t in answering better. It’s in working better. The weak way to use it is to ask for answers. The strong way is to build a system with context, memory, subagents, verification, and continuous execution. Those who use the model only as chat will get better answers. Those who use it as a work system will gain a real advantage.
— conclusion of the practical guide that gave rise to track 2 (not an official Anthropic quote)
In one sentence: chat gives better answers; a workflow system gives you a real advantage — and you build the difference, not the model.
How Tracks 1 and 2 connect
Track 2 is applied—it grew out of a practical guide, not the official docs. But almost every tip in it has a "technical side" already covered in Track 1, with direct support from Anthropic's documentation. If a tip from Track 2 made sense to you and you want to understand the mechanism behind it, the map below takes you straight to the right module.
Give the why
Tip 2 (explain the reason) is the applied/product version of module 1.2, which is more technical/API-level (category reasoning_extraction).
Effort and depth
Tip 3 (control the depth) is the practical equivalent of the official effort levels—low/medium/high/xhigh/max.
Clear boundaries
Tip 4 (set limits before acting) is the same discipline of “say what not to do” as official negative prompting.
Prove you’re done
Tip 5 (real verification) is literally the same evidence requirement as module 1.4—just rephrased in product language.
Delegation to subagents
Module 1.7 teaches as how to delegate technically; tip 6 (specialist subagents) teaches you what roles use in practice.
File-based memory
Tip 7 (project memory) follows the same golden rule as module 1.7: one lesson per file, update instead of duplicate.
Anti-overplanning
Tip 8 (continue when you already have enough information) is the same lesson from module 1.4 about not ending with a promise.
Defensive review
Tip 10 (ask for a defensive review) is the practical strategy for avoiding the refusal cyber explained in module 1.6.
Verification by subagent
Tip 11 (verification loops) uses the same mechanism as module 1.4: a fresh verifier, with new context, checking against the specification.
📄 Official support (track 1)
About actual verification (2.3 ↔ 1.4): "In Anthropic's testing, this nearly eliminated fabricated status reports even on tasks designed to elicit them."
About project memory (2.4 ↔ 1.7): "Store one lesson per file with a one-line summary at the top... update an existing note rather than creating a duplicate; delete notes that turn out to be wrong."
In one sentence: nearly every practical tip in Track 2 has a corresponding technical/official module in Track 1 — use this map to go deeper when you need to.
Final checklist for the entire course
The final cheat sheet: it combines the 6 techniques from the video + 5 extras from the bonus in track 1 (modules 1.2 to 1.7) with the 12 tips from track 2 (modules 2.1 to 2.6). It’s the densest reference in the course — paste it into your CLAUDE.md and come back here whenever you’re about to start a long, important session with Fable 5.
CHEAT SHEET COMPLETO — Prompting Claude Fable 5 (trilha 1 + trilha 2) — TRILHA 1 — Prompting Claude Fable 5 (módulos 1.2–1.7) — [1.2] Dê o porquê — contexto de intenção antes da tarefa [1.2] Não peça pra mostrar o raciocínio (evita a recusa reasoning_extraction) [1.3] Negative prompting — diga o que NÃO fazer, sem refactor/limpeza não pedidos [1.4] Deixe agir quando já tem informação suficiente; pause só pro que só o humano decide [1.4] Prove que terminou — audite cada alegação contra evidência de ferramenta [1.5] Diga menos, não mais — lidere com o resultado, corte o que não muda a próxima ação [1.6] Calibre o esforço (low/medium/high/xhigh/max) — baixo no Fable 5 ainda rende bem [1.6] Trate stop_reason:"refusal" como HTTP 200, não como erro; configure fallback pro Opus [1.7] Delegue subtarefas independentes a subagents; prefira comunicação assíncrona [1.7] Guarde memória em arquivo — uma lição por arquivo, resumo de 1 linha no topo [1.7] send_to_user pra deliverable parcial/update com números — só com instrução explícita [1.7] "Continue" resolve o early stopping raro; reforce quando parar vs. quando agir [1.7] Refatore skills antigas — comece no topo da dificuldade; verifique com subagent fresco — TRILHA 2 — Fable 5 como Agente de Trabalho (módulos 2.1–2.6) — [2.1] Use o Fable 5 pra problemas difíceis — arquitetura, risco, plano, auditoria [2.1] Explique o motivo do pedido — objetivo, público e resultado esperado [2.2] Controle a profundidade — baixa/média/alta/máxima conforme a complexidade [2.2] Defina limites antes de deixar o agente agir [2.3] Exija verificação real — evidência de ferramenta, nunca "pronto" sem prova [2.3] Use subagentes por especialidade — estratégia, produto, comunicação, tecnologia, cético [2.4] Crie memória de projeto — um arquivo por tema, atualize em vez de duplicar [2.4] Mande o agente continuar quando já houver informação suficiente [2.5] Use visão pra revisar interfaces, prints e criativos [2.5] Peça revisão defensiva pra segurança — nunca exploração/ataque [2.6] Crie loops com verificação — foco, critério, teste, aprendizado [2.6] Use memória externa pra trabalhos longos — não dependa só do histórico do chat
💡 Practical tip
Before a long session or a new project, read this list from top to bottom and mentally check which lines are already in your prompt/CLAUDE.md. The missing ones are the most likely reason the session delivers less than it should.
In one sentence: 13 techniques from track 1 + 12 tips from track 2, one list — the reference to keep after finishing the course.
🎓 Module Summary
This concludes the course:
You went through the 14 modules of "Prompting Claude Fable 5"—the 7 parts of track 1 (the grammar of how to talk to Fable 5) and the 7 parts of track 2 (the manual for how to work with it). Return to the topic 6 checklist whenever you start a long session, and to the master prompt in topic 2 whenever you start a new project.