Summary

FableCodex enhances the Codex coding agent by integrating Fable-inspired, evidence-based workflow gates. It targets users who require rigorous process for complex tasks, offering improved discipline and verifiable completion over raw model capability.

How It Works

This Codex plugin overlays a structured workflow onto agent operations. It enforces pre-action inspection, goal tracking with evidence checkpoints, and a mandatory findings gate before completion. The approach prioritizes procedural discipline and verifiable outcomes, leveraging Codex-native tools rather than altering model weights or core intelligence.

Quick Start & Requirements

Install the stable release via codex plugin marketplace add baskduf/FableCodex --ref v0.5.1 and codex plugin add codex-fable5@fablecodex. Restart Codex to activate. For optional provider bridging (model routing), valid Anthropic access and an OpenAI-compatible gateway (e.g., LiteLLM) are required. Local state is managed in .codex-fable5/.

Highlighted Details

• Goal Ledger: Manages multi-step tasks locally (goals.json), requiring evidence for each completed goal.
• Findings Gate: Tracks and necessitates resolution of review issues (findings.json) before task finalization.
• Coverage Accounting: Assesses adherence to source-heading guidelines against local documentation (CLAUDE-FABLE-5.md).
• Codex-Native: Adapts workflow concepts from Fable-style models into native Codex operations.

Maintenance & Community

The project acknowledges contributions from a community of developers. Specific community channels (Discord, Slack) or a public roadmap are not detailed in the README. It has been recognized in sickn33/antigravity-awesome-skills#686.

Licensing & Compatibility

Licensed under AGPL-3.0-or-later. This strong copyleft license may impose restrictions on integration with proprietary or closed-source software, requiring derivative works to also be open-sourced under the same terms.

Limitations & Caveats

FableCodex does not modify underlying model weights, context length, or safety systems; it solely provides workflow guidance. It is unsuitable for simple queries or tasks where process overhead exceeds work value. Model access must be provided externally.