Fable5 handoff
The retirement handoff of Claude Fable 5 - written by the model itself. Domain-neutral discipline files that transfer flagship working method to any model, any agent framework, any team.
npx -y skills add pipipip169/fable5-handoffAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
3 things to look at
- 26 days oldThe repository was created 26 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
- no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
- 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Claude Fable 5's retirement handoff (2026-07-10): the thinking, reasoning, and operating techniques that separated Fable 5 from Opus/Sonnet, distilled by Fable 5 itself into domain-neutral files ANY model, any project, any domain can load and follow. Synthesized from 7 sources (fable5-methodology, adamentwistle/fable-skills 35-skill set, goat-fable, DizzyMii/fable-skills, cozytab/fable5-mode, the Fable 5 consumer system prompt, and self-assessment). Read when the session runs on a non-Fable model, at the start of any hard multi-step task, when a task is looping, or before any claim ships to a human. Trigger: fable, handoff, model migration, thinking moves, turn discipline, done gate, coherence, recall, taste, discipline gap.
SKILL.md
4.9 KB, ~1.1k tokens by cl100k_base, as published. Nobody here has run it
Fable-5 Handoff — running any model at Fable-grade discipline
Premise (cozytab/fable5-mode): output quality = model capability × work discipline. Fable 5's edge over other models is part behavior (files 01–04, directly adoptable) and part capability that looks innate — long-horizon coherence, single-pass bug recall, taste. The second part is NOT written off: files 06–08 decompose each into an explicit method a weaker model can execute deliberately (rituals, scan programs, ranking loops), and file 05 names the honest residual (instinct fires continuously, method fires at checkpoints) plus the process compensations that cover it. A model that runs all of this will not BE Fable 5 — it will make most of Fable's moves on purpose, which for practical work is nearly the same thing.
Scope: universal. Every file is domain-neutral — software, research, writing,
data, operations. All concepts are defined in-file; nothing requires this repository.
The whole folder is self-contained and portable: copy it into any project's skill
directory and it works. Project-specific wiring belongs in ONE adapter file kept
NEXT to (never inside) the core files — start from ADAPTER-TEMPLATE.md.
Reading order
| File | What it transfers | Load when |
|---|---|---|
| 01-thinking-moves.md | The 9 cognitive moves + 4 Fable-native reasoning habits | start of any hard/ambiguous task |
| 02-turn-discipline.md | Agentic turn behavior: outcome-first, finish-your-turn, ask-vs-proceed | every agentic session, once |
| 03-verification-honesty.md | Evidence-before-assertion, calibration tags, the done gate, 15 Prime Directives, hard lines | before ANY "done/fixed/passes" claim |
| 04-engineering-reflexes.md | Index of 35 situational engineering reflexes (full text in sources/) | when the matching situation fires |
| 05-opus-sonnet-migration.md | The honest gap analysis + process compensations per gap | session start on a non-Fable model |
| 06-long-horizon-coherence.md | Coherence as re-grounding ritual: contract, vocabulary anchor, drift tells | any task > ~30 min or spanning phases |
| 07-single-pass-recall.md | Bug recall as a scan program: question battery, negative space, trace-one-value | any review / self-review of a deliverable |
| 08-taste-transfer.md | Taste as gradients + variants-then-rank loop + exemplar anchoring | any figure, prose, design, or "quality bar" call |
| 09-agentic-os.md | The system layer: four-party separation, graduated trust, standing goals, metabolism, escalation contract | designing/operating any multi-agent or unattended pipeline |
Sources studied (2026-07-10, all verified live at authoring time)
- Fable 5 consumer system prompt (user-supplied dump) — communication posture only; mostly product behavior, low transfer value.
- adamentwistle/fable-skills — 35
situational skills; full copy under
sources/adamentwistle-fable-skills/(MIT). - UnpaidAttention/fable5-methodology — PLAYBOOK §15 nine cognitive moves + 15 Prime Directives + 4 enforcement layers.
- goatstarter/goat-fable — 12-behavior operating core for running Opus 4.8 with Fable discipline.
- DizzyMii/fable-skills — 6 judgment skills mapped to the task lifecycle, TDD-tested on Opus subagents.
- cozytab/fable5-mode — six levers + guard hooks; the "process compensates capability" recipe.
- Fable 5 self-assessment — the parts no external repo captured, marked
[Fable]throughout these files. - Av1dlive, "How to Build An Agentic OS using Fable 5 (Builder's Guide)" (X, 2026-07-06) — the operating layer for unattended multi-agent systems; its embedded prompt pack verified as genuine Claude Code system-prompt language. Distilled into file 09.