get-fableGitHub

Your model is capable of more than the way you run it

get-fable adds frontier-style execution discipline around AI coding agents with specs, persistent task state, lifecycle hooks, reusable skills, failure handling, and verification

View on GitHub Read the thesis
The weights stay the sameThe working conditions do not
01 / THE QUESTION

Why do frontier models feel different?

Raw model capability is part of the answer. It is not the whole answer

A strong agent experience also depends on what happens around the model: how the task is framed, what context survives between turns, when implementation starts, how repeated failures change behavior, which skills are available, and what counts as proof that the work is finished

get-fable focuses on that part: the execution environment around the model. It does not change model weights and it does not claim equivalence with Claude Fable 5, Claude Mythos 5, GPT-5.6 Sol, or any other frontier model

How much of frontier-style execution can we bring to the coding agents we already use?

Claude Fable 5 / Mythos 5 GPT-5.6 Sol
02 / THE DIFFERENCE

Same model. Different run

Ordinary run
  1. 01Vague instruction

    The task starts before success is made explicit

  2. 02Context drift

    Important state lives only inside the conversation

  3. 03Blind retries

    The same class of failure receives more of the same response

  4. 04Looks done

    Completion is inferred from output rather than evidence

With get-fable
  1. 01Spec before work

    Goal, approach, checks, dependencies, and decisions stay visible

  2. 02Persistent task state

    The ledger keeps execution state outside the chat

  3. 03Failure changes strategy

    Repeated command failures become a different problem to diagnose

  4. 04Evidence before close

    Unfinished work and missing proof can block completion

03 / THE DISCIPLINE

Six ways to change the run without changing the model

Not more prompting. Better execution conditions around the work

  1. 01

    Plan before implementation

    Turn intent into a concrete project spec before expensive execution starts

  2. 02

    Keep task state outside the chat

    Use durable files for tasks, progress, decisions, and evidence instead of trusting conversation history alone

  3. 03

    Carry working rules across turns

    Reintroduce the active project state and execution rules when a new agent session starts

  4. 04

    React differently when failures repeat

    Move from retrying commands to identifying the class of failure when the same path keeps breaking

  5. 05

    Reuse skills and agent instructions

    Keep useful operating knowledge available as inspectable assets instead of recreating it in every prompt

  6. 06

    Require evidence before completion

    Make “done” answer to unresolved tasks and observable proof, not confidence or presentation quality

04 / THE RUN

A better run has a sequence

  1. 01Promptintent
  2. 02Specdefinition
  3. 03Ledgerstate
  4. 04Hooksgates
  5. 05Workexecution
  6. 06Verifyevidence
05 / TRY IT

Inspect first. Install second

get-fable is open source and the hooks are plain files you can review before adding them to your environment

Read the full quick start
terminal
git clone https://github.com/imMamdouhaboammar/get-fable.git
cd get-fable

bun ./bin/get-fable.js status
bun ./bin/get-fable.js assets
bun ./bin/get-fable.js install

06 / THE BOUNDARY

Strong claims are easy. Inspectable claims are better

What is implemented

  • Claude CodeSkill, project rules, and four lifecycle hooks
  • Antigravity / Gemini configPlugin package, skill files, and rules under the documented config target
  • Agent KernelRules copied when the existing Agent Kernel directory is present
  • Request proxyOpenAI-style request enrichment with one configured upstream URL

What get-fable does not claim

  • No model replacementIt does not reproduce a proprietary model or private service
  • No universal installerIt does not automatically configure every IDE, provider, or coding agent
  • No correctness guaranteeBetter process does not eliminate hallucinations or execution risk
  • No hardened public gatewayThe current request proxy is a development utility and needs an external access-control boundary before untrusted exposure
07 / OPEN SOURCE

Read the code. Read the hooks. Challenge the thesis

get-fable is an independent community project. References to model, company, and upstream project names are descriptive only and do not imply affiliation, endorsement, or sponsorship

RepositoryMIT LicenseThird-party notices

The model matters.
The way you make it work matters too

Get the code