imMamdouhaboammar/get-fable

imMamdouhaboammar/get-fable

DeepSeek Harness 的模块化执行纪律与自主管家:覆盖 44 项规范技能的循证路由、持久化状态规划、验证门禁、失败恢复与交互式 Web 看板。

工作流与自动化agentagentic-aiai
get-fable rabbit mascot

get-fable

What if the coding agent you already use could work more like Claude Fable?

Same model. Better working habits. A more accountable coding loop.

Quick install

Add the open-source Skills to your coding agent or project:

npx skills add imMamdouhaboammar/get-fable

Install the CLI separately with your preferred package manager:

# Recommended: Bun 1.3+
bun add -g get-fable

# Alternative: npm (the CLI still requires Bun to run)
npm install -g get-fable

Or use the shell installer after reviewing it:

curl -fsSL https://raw.githubusercontent.com/imMamdouhaboammar/get-fable/master/install.sh | sh

Then, from the root of your own project:

get-fable --version
get-fable init
get-fable install all
get-fable skills install all
get-fable doctor --json-v1
get-fable lint

install all targets supported coding hosts, while skills install all deploys the available Skills. Both can change local or global agent settings. Check what is installed and back up custom host configurations first. Run init only inside the project you intend to set up.

Ask your agent to install and configure get-fable

Copy this into Codex, Claude Code, Cursor, Gemini CLI, or the coding agent you use:

Set up the latest public get-fable release for this project.

Read and follow the install-and-init-get-fable onboarding playbook:
https://github.com/imMamdouhaboammar/get-fable/blob/master/docs/install-and-init-get-fable.md

1. Inspect the OS, repo root, available agent hosts, existing configurations,
   and Bun/Git/Python versions. Protect existing settings and uncommitted work.
2. Install or update the get-fable CLI via bun add -g get-fable, or use npm
   if Bun's package manager is not available. Verify the installed version.
3. Install the public Skills via npx skills add imMamdouhaboammar/get-fable.
4. In this project's root run get-fable init, then configure supported hosts
   using get-fable install all and Skills using get-fable skills install all.
5. Check get-fable status --json-v1, doctor --json-v1, and lint. Run a safe
   routing smoke test without changing application code.
6. Explain the daily workflow in my preferred language and provide a concise
   report of versions, installed hosts, Skills, checks, skipped steps, and fixes.

Before global toolchain changes, shell changes, git hook replacements, or
fable-eco capability provisioning, show the plan and request my consent.
Never overwrite custom configuration or handle credentials without approval.
Do not claim success unless the final checks actually pass.

Full copy-ready agent prompt with acceptance criteria →


The problem is not always the model

You ask an AI coding agent to fix a bug. It jumps into an edit, produces something plausible, runs a test, and declares victory. An hour later, you discover it tested the wrong thing, or changed more than the task needed, or forgot the original decision.

Maybe a stronger model would have avoided it. But some of these failures come from how the work is organized.

A useful engineering partner needs to know when to investigate, when to write a failing test, when to get the current API contract, and when to admit the evidence is not good enough yet.

That is the idea behind get-fable.

It is a portable working method for the coding agent you already use. The model still writes the code. The Skills and CLI help it make better decisions about the work around the code.

A bug fix, with a little more discipline

Without an explicit lifecycleWith get-fable
Start editing from a guessLocate the real execution path first
Fix the apparent symptomReproduce the failure at a meaningful test boundary
Lose the plan across sessionsKeep work cards and decisions in .fable/
Repeat the same failed patchPause and revisit the diagnosis
Report a green test from before the last changeCollect fresh evidence against the current files
Announce completion on confidenceCheck acceptance criteria before closing

Get-fable does not guarantee correct code. It gives a coding agent a more explicit process, and gives you better places to check whether it followed that process.

One workflow, several specialized Skills

The public source currently includes 41 installable community Skills across 8 packs. The adapters cover 32 host targets, with different integration levels depending on the platform. A rule file, an installed Skill, and a lifecycle hook are different levels of support; they are not interchangeable.

StageWhat the agent should establishExample public Skills
DiscoverWhat actually runs, and where?get-fable, fable-discover, fable-research
PlanWhat is in scope, and what proves success?fable-plan, fable-architecture
BuildCan we show the gap before changing production code?fable-tdd, fable-execute
VerifyWhat could still disprove the claim?fable-verify, fable-review, fable-security
RecoverWhy did the previous approach fail?fable-recover
DeliverIs the artifact ready, and can someone else resume?fable-release, fable-handoff

The core lifecycle is useful on its own. Extra packs support architecture, security, visualization, memory, and other specialized work. See the public catalog for what is included in this repository.

Keep state outside the conversation

The CLI can initialize project-local tracking files:

.fable/
  state.json      # lifecycle and evidence state
  LEDGER.md       # work cards and acceptance checks
  PROGRESS.md     # what changed and why

The exact files and host integration depend on the installed CLI version. A Skills-only setup works without requiring the CLI to be present, but it cannot provide CLI-backed state checks and hooks.

That distinction is intentional: you can start small, inspect the method, and add the runtime support when you need it.

Works with the agent you already use

Get-fable has integration targets for Claude Code, Codex, Gemini/Antigravity, Cursor, OpenCode, Grok, and other tools. Support varies by host: some receive Skills, others rule files or lifecycle hooks. Use get-fable install all for supported host setup, or select a host individually, for example get-fable install codex.

Run the CLI's doctor to inspect your actual environment. A listed target does not imply every feature works on every host.

Community today. A deeper MCP harness in development.

Available now: Community / OSS SkillsIn development: Get-Fable Pro MCP
Open-source Skills, readable playbooks, documented lifecycle, and an optional CLI package. Licensed under AGPL-3.0-or-later.A planned premium MCP-based engineering harness with deeper agent and tool integration. No release date, specification, price, or performance claim is being made yet.

The public Skills remain open source. Pro MCP is a future product direction, not a requirement for using the community edition. We want to build a remarkably capable harness, but we'll judge it by demonstrable behavior when it is released.

What it is, and what it is not

  • It is an inspectable set of engineering Skills and a separately distributed CLI.
  • It does not change a model's weights, grant capabilities its host does not have, or make verification automatic on every platform.
  • It does not promise zero mistakes, zero hallucinations, perfect autonomy, or a passing CI run for your project.
  • It does encourage evidence you can inspect: tests, logs, state, diffs, and clear reasons to stop or continue.

If you are using the npm CLI, check its package version independently of the public GitHub checkout. The source-only public repository is not a runnable replacement for the npm CLI distribution.

Documentation

Looking for...Start here
Quick setup and prerequisitesInstallation
Complete guided onboardingInstall and initialize
Copy/paste agent instructionAgent setup prompt
Available community SkillsSkill catalog
Host-specific integration levelsHost matrix
Day-to-day commandsUsage
Architecture and evidenceArchitecture · Evidence
Contributing and responsible disclosureContributing · Security

Contributing

Issues, documentation corrections, tests, and contributions to public Skills are welcome. Please keep the contribution guidelines and security policy in mind. Internal engineering assets, credentials, private evaluations, and unpublished commercial material are not community distribution material.

License: AGPL-3.0-or-later. get-fable is an independent community project and is not affiliated with Anthropic or Claude Fable.