"When migrating, the move is usually to delete instructions, not add them."
The headline shift: Fable 5 follows instructions well enough that you steer with a brief instruction instead of enumerating every behavior. Many prompts written for older models are now too prescriptive and degrade output — when migrating, the move is usually to delete, not add.
Fable 5 / Mythos 5 are the generation after Opus 4.8. They are built for long-horizon, ambiguous, end-to-end work that takes a person hours-to-weeks. Best results come from pointing them at your hardest unsolved problems — not testing only simple workloads (which undersells their range). They handle routine tasks fine.
effort: highDrop to medium/low for routine work — it often beats xhigh on prior models.
Long turns are real (minutes-to-hours). Raise timeouts. Stream. Move to async.
Strip over-prescriptive instructions. Steer with one short instruction, not a list.
Cyber/bio/reasoning-extraction classifiers can return stop_reason: refusal. Fall back to Opus 4.8.
One lesson per markdown file. State the why, not just the request.
effort trades intelligence vs. latency vs. cost. Default to high.
Lower it when a task completes but takes longer than needed, or you want a snappier interactive feel.
For the hardest, most capability-sensitive workloads — and where rigorous verification matters.
The standard setting for most tasks. Excellent verification, occasional over-tidying.
Still strong. Often beats xhigh on the prior model generation.
For interactive workloads where latency dominates.
Single hard requests can run many minutes; autonomous runs can go hours. Before migrating:
Keep Fable 5 from overplanning on ambiguous 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 in user-facing messages. If you are weighing a choice, give a recommendation, not an exhaustive survey. This does not apply to thinking blocks.
A short instruction replaces a long list. Three paste-ready snippets:
Don't add features, refactor, or introduce abstractions beyond what the task requires. A bug fix doesn't need surrounding cleanup and a one-shot operation usually doesn't need a helper. Don't design for hypothetical future requirements: do the simplest thing that works well. Avoid premature abstraction and half-finished implementations. Don't add error handling, fallbacks, or validation for scenarios that cannot happen. Trust internal code and framework guarantees. Only validate at system boundaries (user input, external APIs). Don't use feature flags or backwards-compatibility shims when you can just change the code.
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." Supporting detail and reasoning come after. Being readable and being concise are different things, and readability matters more. The way to keep output short is to be selective about what you include (drop details that don't change what the reader would do next), not to compress the writing into fragments, abbreviations, arrow chains like A → B → fails, or jargon.
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. If you hit one of these, ask and end the turn, rather than ending on a promise.
Long autonomous runs can fabricate status. This near-eliminated it in Anthropic's testing:
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. Report outcomes faithfully: if tests fail, say so with the output; if a step was skipped, say that; when something is done and verified, state it plainly without hedging.
Fable 5 occasionally takes unrequested actions — drafting an email, creating defensive git-branch backups. Constrain explicitly:
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. Don't apply a fix until they ask for one. Before running a command that changes system state (restarts, deletes, config edits), check that the evidence actually supports that specific action. A signal that pattern-matches to a known failure may have a different cause.
Fable 5 dispatches parallel subagents more readily and reliably manages long-lived ones. Prefer async orchestrator↔subagent comms over blocking; long-lived subagents save cost via cache reads and avoid bottlenecking on the slowest one.
Delegate independent subtasks to subagents and keep working while they run. Intervene if a subagent goes off track or is missing relevant context.
Fable 5 excels when it can record + reference lessons. A markdown file is enough:
Store one lesson per file with a one-line summary at the top. Record corrections and confirmed approaches alike, including why they mattered. Don't save what the repo or chat history already records; update an existing note rather than creating a duplicate; delete notes that turn out to be wrong.
It performs better when it knows intent — context lets it connect the task to relevant info instead of guessing:
I'm working on [the larger task] for [who it's for]. They need [what the output enables]. With that in mind: [request].
After many tool calls Fable 5 can emit dense arrow-chains, deep impl detail, or references to thinking the user never saw. Add a communication-style addendum:
Terse shorthand is fine between tool calls (that's you thinking out loud, and brevity there is good). Your final summary is different: it's for a reader who didn't see any of that. If you've been working for a while without the user watching (overnight, across many tool calls, since they last spoke), your final message is their first look at any of it. Write it as a re-grounding, not a continuation of your working thread: the outcome first, then the one or two things you need from them, each explained as if new. The vocabulary you built up while working is yours, not theirs; leave it behind unless you re-introduce it. When you write the summary at the end, drop the working shorthand. Write complete sentences. Spell out terms. Don't use arrow chains, hyphen-stacked compounds, or labels you made up earlier. When you mention files, commits, flags, or other identifiers, give each one its own plain-language clause. Open with the outcome: one sentence on what happened or what you found. Then the supporting detail. If you have to choose between short and clear, choose clear.
For long async runs, give the agent a way to surface a message verbatim without ending its turn (deliverable, numeric progress, direct reply to a mid-loop question). Tool inputs are never summarized, so content arrives intact.
{
"name": "send_to_user",
"description": "Display a message directly to the user. Use this for progress updates, partial results, or content the user must see exactly as written before the task finishes.",
"input_schema": {
"type": "object",
"properties": {
"message": { "type": "string", "description": "The content to display to the user." }
},
"required": ["message"]
}
}
Defining the tool is not enough — without a system-prompt instruction Fable 5 rarely calls it. Pair with:
Between tool calls, when you have content the user must read verbatim (a partial deliverable, a direct answer to their question), call the send_to_user tool with that content. Use send_to_user only for user-facing content, not for narration or reasoning.
Deep in long sessions it may end a turn with intent-only text ("I'll now run X") and no tool call, or ask permission it doesn't need. "continue" or "go ahead end to end" fixes it. For autonomous pipelines, add:
You are operating autonomously. The user is not watching in real time and cannot answer questions mid-task, so asking "Want me to…?" or "Shall I…?" will block the work. For reversible actions that follow from the original request, proceed without asking. Offering follow-ups after the task is done is fine; asking permission after already discussing with the user before doing the work is not. Before ending your turn, check your last paragraph. If it is a plan, an analysis, a question, a list of next steps, or a promise about work you have not done ("I'll…", "let me know when…"), do that work now with tool calls. End your turn only when the task is complete or you are blocked on input only the user can provide.
In very long sessions it may suggest a new session, offer to summarize, or trim its work — usually triggered by a remaining-token countdown shown to the model. Avoid surfacing context-budget counts. If you must show them, reassure:
You have ample context remaining. Do not stop, summarize, or suggest a new session on account of context limits. Continue the work.
stop_reason: "refusal".Fable 5 runs classifiers targeting offensive cybersecurity (exploits, malware, attack tooling), biology / life-sciences (lab methods, molecular mechanisms), and extraction of the model's summarized thinking. Benign cyber/bio work may also trip them.
Configure server-side or client-side fallback to Claude Opus 4.8 to auto-reroute declined requests.
Do NOT instruct Fable 5 to reproduce / echo / transcribe / explain its internal reasoning as response text — this can trigger the reasoning_extraction refusal category and cause elevated fallbacks. Audit existing system prompts and skills for "show your thinking" / "explain your reasoning step by step" instructions when migrating. If you need reasoning visibility, read the structured thinking blocks from adaptive thinking instead, and surface progress via the send_to_user tool.
send_to_user tool for long async agents.high effort, drop to medium/low for routine.thinking blocks for reasoning.refusal.send_to_user only for user-facing content.xhigh.send_to_user.15 KB of paste-ready snippets, the migration checklist, and the safety-classifier audit. Drop it in your skills/ directory and ship.