fable-orchestrator README

The project's README from github.com/codejunkie99/fable-orchestrator. As one markdown file: README.md.

Fable orchestrator

Fable is a small, local-first routing skill for Codex. Claude Fable 5.1 plans and adjudicates; it does not write code or own the workspace. Codex remains the runtime and delegates bounded implementation work to OpenCode Go agents:

The skill consumes the callable opencode-go/ and opencode-go-responses/ agents supplied by Codex Router. It does not ship a proxy, dashboard, model catalog, credential store, or API key. Configure OpenCode Go once in the router, then restart Codex after changing provider or agent definitions.

Fable orchestrator: planning, implementation, and verification

Repository layout

skill/fable/
├── SKILL.md
├── agents/openai.yaml
└── scripts/ask_fable.sh
assets/fable-orchestrator.svg
install.sh
tests/test_skill.sh

The three files under skill/fable/ are the installable skill. ask_fable.sh is executable and invokes Claude Code's local fable alias with no session persistence. The packet passed to it must contain decisions and workspace facts only; never put credentials in a packet.

Install

From this repository:

./install.sh --dry-run
./install.sh --copy

--dry-run prints the exact destination and copy operations without creating files. --copy installs to ~/.codex/skills/fable, creates only the required directories, and is safe to run again. Use --target DIR to select another skills directory:

./install.sh --copy --target "$PWD/.local/codex/skills"

The installer reads only this repository and the destination path. It never reads, creates, or modifies credentials. Start a new Codex task after changing the provider or agent definitions.

Use

Invoke the skill with an objective:

$fable build the feature

Fable returns a bounded graph. Codex validates the graph, starts ready workers in parallel when useful, collects their evidence, verifies the result, and asks Fable to adjudicate when the task needs another decision. Every implementation node must use GPT-5.6 Luna or DeepSeek V4 Flash. If neither allowed OpenCode Go route is callable, the workflow reports the blocker instead of inventing a model.

Test

Run the repository's dependency-light checks:

tests/test_skill.sh

The test checks shell syntax, the copied source files, required routing strings, basic YAML structure, optional xmllint XML validation, SVG safety constraints, a temporary-home dry run, an idempotent copy, and common credential-shaped strings. It does not require PyYAML or a live Claude login.

License

MIT. See LICENSE.

fable-orchestrator (codejunkie99/fable-orchestrator)

Searching Database…

616 ·GitHub
Fable-orchestrator
这是什么？Fable 是一个本地优先的 Codex 路由技能，负责规划和裁决任务，把范围有限的实现工作委派给 OpenCode Go 智能体——日常工作用 GPT-5.6 Luna，循环和高吞吐迭代用 DeepSeek V4 Flash。
这是一个独立应用还是现有工具的技能？它是一个 Codex 技能，不是独立应用：skill/fable/ 下的三个文件会被复制进 Codex 的 skills 文件夹，Codex 仍是运行时，Fable 只负责规划和裁决，不写代码。
使用 Fable 之前需要先配置什么？你需要配置好 Codex，在 Codex Router 中配置 OpenCode Go 智能体（GPT-5.6 Luna 和 DeepSeek V4 Flash），另外还要有 Claude Code 的本地 fable 别名，askfable.sh 会调用它。
如果所需的模型不可用会怎样？如果 GPT-5.6 Luna 和 DeepSeek V4 Flash 都无法通过 OpenCode Go 调用，Fable 会报告一个阻塞问题而不是替换成别的模型；另外，更改提供商或智能体定义后必须重启 Codex。
Fable 会碰我的凭据吗？不会——该技能及其安装器从不读取、创建或修改凭据，传给它的任何数据包都只能包含决策和工作区事实。
为什么有两个不同的实现模型而不是一个？GPT-5.6 Luna 负责常规实现，DeepSeek V4 Flash 专用于循环、重复迭代和高吞吐工作，这样 Fable 可以把每项任务路由到最适合它的模型。
测试怎么跑？仓库的轻依赖测试套件会检查 shell 语法、复制过来的源文件、必需的路由字符串、基本 YAML 结构、可选的 XML/SVG 校验、临时主目录试运行、幂等复制，并扫描形似凭据的字符串，全程不需要 PyYAML 或真实的 Claude 登录。
Fable 如何判断任务何时需要再次决策？Codex 校验 Fable 返回的图，并行启动就绪的 worker，收集它们的证据并验证结果，只有当任务需要再次决策时才回头调用 Fable 进行裁决。
What are similar projects?pstack-claude, council-of-high-intelligence, ordewell, agents, ruflo, foreman, pstack, hyperfide
Which topics is it in?bash, agent-skill, gpt, harness, opencode, orchestration, agent-orchestration, cli-tool, skill, shell, deepseek, agentic-ai
Creator
codejunkie99
Size
<0.1 MB
Stars
616
Updated
CommitSep 04
Age
1mo
Commits · 90 Days