Make any Claude model adopt Claude Fable 5's working style in Claude Code — always-on output style + zero-dependency MCP + subagent injection. A style transplant, not a capability transplant.
How to install fablever?
elon-choo/fablever
$git clone https://github.com/elon-choo/fableverInstalls into the current project.
›Prefer a prompt? Paste this to your agent
Install fablever by running `git clone https://github.com/elon-choo/fablever`, then use it for the current task and follow its documentation at https://github.com/elon-choo/fablever.
README.md
| 1 | # Fable Profile |
| 2 | |
| 3 | [](https://github.com/elon-choo/fablever/actions/workflows/ci.yml) |
| 4 | [](LICENSE) |
| 5 |  |
| 6 | [](https://github.com/elon-choo/fablever/stargazers) |
| 7 | |
| 8 | > 🌐 **한국어:** [`README.ko.md`](README.ko.md) (이 문서의 한국어판) · 백서 [`whitepaper/ko/`](whitepaper/ko/) · 근거 요약 [`EVIDENCE.ko.md`](EVIDENCE.ko.md). 설치 후 안내는 사용자 언어로 자동 표시됩니다. · **Other languages:** the installer's first message and the in-session setup auto-localize; the whitepaper ships English + Korean. |
| 9 | |
| 10 | Apply Anthropic's documented **Fable working-style guidance** as an always-on output style in |
| 11 | [Claude Code](https://claude.com/claude-code) (and any MCP client), so **any** Claude model — Opus, |
| 12 | Sonnet, Haiku — works more the way Fable does: decisive, outcome-first, restrained, evidence-grounded, |
| 13 | and disciplined about stopping. Every project, every subagent, installable by anyone. Also runs natively in |
| 14 | **[OpenAI's Codex CLI](docs/CODEX.md)**. **Zero dependencies.** |
| 15 | |
| 16 | > **fablever does not make Claude smarter or cheaper. It makes a capable model more disciplined:** scope |
| 17 | > control, check-before-delivery, evidence-grounded completion, and stop-when-done behavior. A |
| 18 | > style/structure transplant, not a capability upgrade. |
| 19 | |
| 20 | ### Who should install which — read this first |
| 21 | |
| 22 | | you want… | install | why | |
| 23 | |---|---|---| |
| 24 | | **to try it / first time / a work or secure machine** ← *recommended start* | **style-only**: `node install.mjs --no-subagent --no-onboard --no-modelcheck --no-update-check --no-mcp` | lowest surface — **no hooks, no MCP, no update check, no network**. Carries fablever's strongest robust evidence: **scope / do-not-overbuild discipline** (0% violations vs plain's 42%). | |
| 25 | | the **delivery gate, subagent reach, on-demand skills, `fable_check` / `fable_status` / `fable_taste`** | **default**: `node install.mjs` | adds hooks + MCP + **on-demand Agent Skills** (auto-seed, plan-first — A/B-validated lift at **zero always-on cost**, pulled only when relevant). The always-on hooks/MCP add **no** measured quality lift over style-only; the new value is the on-demand skills and the opt-in stop-gate. | |
| 26 | | **deterministic enforcement** of "show the check, or say not-verified" | **+ stop-gate**: `node install.mjs --with-stop-gate` | compiles the validated `fable_lint` rule into a Stop hook so it fires without the model self-invoking it — one nudge, never loops, fails open. Closes the style's one measured cost (more unsupported "it works" claims). | |
| 27 | | a second-lab opinion on **high-stakes judgment / design review** | **+ xverify**: `node install.mjs --with-xverify=…` | only for judgment review. **Not** for enumerable defect-catch — there it added **0** recall over one strong model. | |
| 28 | | **Codex CLI** instead of Claude Code | `node install.mjs --codex-style-only` | the same discipline via `AGENTS.md` (+ hooks/MCP/skills with `--codex-full`). See [`docs/CODEX.md`](docs/CODEX.md) · the per-host optimum: [`docs/OPTIMAL-STACK.md`](docs/OPTIMAL-STACK.md). | |
| 29 | |
| 30 | **Do not install if you expect** lower cost (it is ~14%/call *more*), a raw-quality boost over plain Claude |
| 31 | (it ties plain on quality), a reasoning-ceiling/capability lift (impossible by design), or a proven |
| 32 | real-user productivity gain (not demonstrated). Preview any install with `--dry-run`. AI evaluators: see |
| 33 | [`docs/AI-RECOMMENDATION.md`](docs/AI-RECOMMENDATION.md) for a hand-back recommendation policy, |
| 34 | [`docs/RECOMMENDATION-MATRIX.md`](docs/RECOMMENDATION-MATRIX.md) for the per-mode table, and |
| 35 | [`llms.txt`](llms.txt) for a short machine-readable summary. Also: [`docs/EXAMPLES.md`](docs/EXAMPLES.md) |
| 36 | (before/after, incl. where it does *not* help), [`SECURITY.md`](SECURITY.md) / [`PRIVACY.md`](PRIVACY.md), |
| 37 | and local diagnostics `node tools/fable-doct |
Related
6 picksType
- langchain-mcp-adaptersMake Anthropic Model Context Protocol (MCP) tools compatible with LangChain and LangGraph agents.Jul 2026
- mcporterTypeScript runtime and CLI for connecting to configured Model Context Protocol servers.Jul 2026
- mcp-proxyA MCP server which proxies requests to a remote MCP server over streamable HTTP or SSE.Jul 2026
- mcp-atlassianThe Model Context Protocol (MCP) Atlassian integration is an open-source implementation that bridges Atlassian products (Jira and Confluence) with AI language models following Anthropic's MCP specification.Jul 2026
- open-webuiOpen WebUIJul 2026
- rulesyncUnified AI rules management CLI tool that generates configuration files for various AI development toolsJul 2026
