--- okf_version: "0.1" title: "ARIA Fable 5 Playbook — OKF v0.1 Knowledge Bundle" description: "A practitioner playbook for using Claude Fable 5 well during its metered window: what to run, how to prompt, effort discipline, and behavior distillation." updated: 2026-07-06 --- # ARIA Fable 5 Playbook An **OKF v0.1 Knowledge Bundle** — a small, self-contained set of markdown *concepts* you can read top-down or hand to an agent. This one is a **practitioner playbook for using Anthropic's Claude Fable 5 well while it is scarce**: Fable 5 is included on paid Claude plans only through a short window and up to a weekly usage cap, after which continued use falls onto metered usage credits. The bundle collects the best AI-community guidance on **what to point it at, how to prompt it, how to spend the limited budget, and how to keep some of its value after the window closes.** Read this index first, then open only the concept you need (progressive disclosure). ## What's inside ### How to prompt it — meta-prompts - [fable5-miessler-meta-prompts](./fable5-miessler-meta-prompts.md) — Daniel Miessler / NetworkChuck: 16 copy-paste **meta-prompts** that aim the strong model at your *deepest* systems (your AI harness, security / attack-surface, and life-direction) so the payoff outlasts the window. ### What to run — use-cases & prompt packs - [fable5-use-cases-peter-yang](./fable5-use-cases-peter-yang.md) — Peter Yang's **five live-demo use cases** (find Fable-worthy work · life/business advice · ship-ready bug audit · plan-then-execute · refactor a skill system), plus effort-dial and cheap-prep / babysit tips. - [fable5-prompt-pack-money-productivity](./fable5-prompt-pack-money-productivity.md) — Greg Isenberg's **money / productivity prompt pack**: AI-judge tournaments, interview-before-build scoping, and pointing one agent at your own large datasets (8 use-case prompts + 5 startup ideas). ### Effort & targeting discipline — spend the budget well - [fable5-carlini-sweep-score-amjad](./fable5-carlini-sweep-score-amjad.md) — Ray Amjad's **Carlini sweep-and-score** method: a *cheap* model rates every file 1–5, discard the 1s–2s, then aim Fable only at the highest **impact × opportunity** targets (and at root causes, not symptoms). ### Build applications — Fable 5 as the engine - [claude-fable5-youtube-content-pipeline](./claude-fable5-youtube-content-pipeline.md) — Jack Roberts's end-to-end **YouTube content pipeline** driven by Claude Code (Fable 5) + Higgsfield + Auphonic; source of the "run **medium** effort, tag the max model only for genuinely intellectual work" token trick. - [fable5-agentic-os-voice-skills-jarvis](./fable5-agentic-os-voice-skills-jarvis.md) — Chase AI's **"Jarvis"**: a voice + router layer over a Claude Code skill architecture (faster-whisper → regex / Haiku / local router → headless execution → Kokoro TTS), with Obsidian as the system of record. ### Behavior distillation — keep the value after the window - [jsonl-behavioral-distillation-playbook](./jsonl-behavioral-distillation-playbook.md) — Mine your local **JSONL session logs**, measure Fable's behavioral fingerprint against your everyday model, and distill the delta into a **playbook** injected at session start — porting Fable's *habits* to the models you keep. ## The cross-cutting theme: effort discipline Every concept here converges on the same scarcity logic — Fable 5 is powerful but metered, so **route deliberately**: prep and triage with cheaper models, run routine execution at lower / medium effort, and reserve the strong tier (and its highest effort) for genuinely hard, high-leverage work. `python okf-cli.py find effort` surfaces the concepts that treat this directly. ## Using this bundle - **With the CLI:** from inside the bundle — `python okf-cli.py index`, `python okf-cli.py find `, `python okf-cli.py read `, `python okf-cli.py selftest`. - **With an agent:** hand it the OKF v0.1 spec plus this directory. Each concept is one markdown file with a `type` in its frontmatter, this `index.md` is the catalog, and `log.md` is the changelog. - Concept names shown in plain text and annotated *(outside this bundle)* refer to related concepts in the wider ARIA-RD vault that were not included here. ## Sources & provenance Extracted from the **ARIA-RD** R&D AI-knowledge vault's Fable 5 cluster on 2026-07-03. Each concept keeps its original `## Sources` / `# Citations` footer with the primary video and the research verified against Anthropic's own announcements. ## Concepts (8) - [claude-fable5-youtube-content-pipeline](./claude-fable5-youtube-content-pipeline.md) — A build-along turning Claude Code on Fable 5 into a faceless YouTube pipeline: deconstruct a channel, script, record your voice, then drive Higgsfield/Auphonic. - [fable5-agentic-os-voice-skills-jarvis](./fable5-agentic-os-voice-skills-jarvis.md) — Chase AI's Jarvis: a web-app layer over Claude Code turning voice into actions via a local STT, router, execution, and TTS pipeline on a skill architecture. - [fable5-carlini-sweep-score-amjad](./fable5-carlini-sweep-score-amjad.md) — Spend the scarce Fable 5 budget well: a cheap model sweeps and scores every file 1–5, discard the low, aim Fable only at high impact × opportunity targets. - [fable5-miessler-meta-prompts](./fable5-miessler-meta-prompts.md) — NetworkChuck and Miessler argue you should aim Fable 5 at your deepest systems — harness, security, life direction — via copy-paste meta-prompts. - [fable5-paper-wargame-kashef](./fable5-paper-wargame-kashef.md) — Instead of asking the scarce Fable 5 window for plans or having it execute builds, have it wargame your hardest projects move-by-move into battle plans a cheaper executor model can run blind after Fable is gone. - [fable5-prompt-pack-money-productivity](./fable5-prompt-pack-money-productivity.md) — Greg Isenberg's playbook of copy-paste prompts pointing Fable 5 at business problems: AI-judge tournaments, interview-before-build scoping, your own datasets. - [fable5-use-cases-peter-yang](./fable5-use-cases-peter-yang.md) — Peter Yang's Fable 5 use cases: find Fable-worthy work, life/business advice, make a project ship-ready, plan a big feature, refactor a skill system. - [jsonl-behavioral-distillation-playbook](./jsonl-behavioral-distillation-playbook.md) — A Claude Code workflow mines JSONL session logs to measure a model's behavioral fingerprint and distill the delta into an injectable playbook. --- --- type: readme source_bundle: "fable5" --- # ARIA Fable 5 Playbook — OKF v0.1 Knowledge Bundle ## What this is A self-contained **OKF v0.1 Knowledge Bundle** collecting AI-community guidance on **using Anthropic's Claude Fable 5 model well** — meta-prompts, concrete use-cases, and effort / execution discipline for a model that is powerful but metered. Eight concepts cover how to prompt Fable 5, what to point it at, how to spend its scarce weekly budget, and how to distill its behavior into the models you keep after the window closes. Start with [`index.md`](./index.md) — the progressive-disclosure catalog. ## How to use - **CLI (stdlib only, no dependencies)** — run from inside this directory: - `python okf-cli.py index` — print the catalog - `python okf-cli.py find ` — ranked keyword search across concepts (e.g. `find effort`) - `python okf-cli.py read ` — print one concept (id = filename minus `.md`) - `python okf-cli.py selftest` — sanity + OKF-conformance check - **With an agent** — hand it the **OKF v0.1 spec** plus this repo / directory. Every concept is one markdown file carrying a `type` in its YAML frontmatter; `index.md` is the catalog and `log.md` the changelog. The agent navigates by reading `index.md`, then opening only the concepts it needs. ## Layout - `index.md` — catalog / entry point (`okf_version: "0.1"`) - `log.md` — changelog - seven concept pages (`*.md`), each with a frontmatter `type` and a preserved `## Sources` / `# Citations` footer - `assets/` — images referenced by the Jarvis concept - `okf-cli.py` — the dependency-free consumer ## Conventions - Links between the seven member concepts are bundle-relative (`./slug.md`). - References to concepts in the wider vault appear as plain text annotated *(outside this bundle)*. ## Provenance Extracted from the **ARIA-RD** R&D AI-knowledge vault on **2026-07-03** (the Fable 5 cluster, 7 concepts). Each concept's original sources — the primary YouTube video and ARIA's complementary research verified against Anthropic's primary announcements — are preserved in its footer.