Analysis updated 2026-05-18
Apply a consistent set of behavioral guidelines to any AI coding assistant.
Read findings.md to see how much these instructions actually change an AI's output.
Use Fable5.md as a template for writing your own operating instructions for a model.
Compare no-instructions, old-instructions, and current-instructions behavior on a real codebase.
| sgup/ai | 5ec1cff/injectrc | elvisun/loss-function-development | |
|---|---|---|---|
| Stars | 65 | 65 | 65 |
| Language | — | C++ | — |
| Setup difficulty | easy | hard | easy |
| Complexity | 1/5 | 4/5 | 3/5 |
| Audience | developer | developer | developer |
Figures from each repo's GitHub metadata at analysis time.
This repository is a small collection of notes and configuration files for getting better results from AI coding assistants. It contains two documents, and the README describes them briefly without going into much technical detail. The first file, Fable5.md, is described as a portable set of instructions written to capture how the Fable 5 AI model approaches coding tasks: how it reasons, decides what to build, and communicates its output. The idea is that you can attach these instructions to any capable AI model to make it behave in a similar way, following principles like verifying before asserting, keeping changes small and reversible, leading with a recommendation when there are multiple options, grounding advice in the actual project rather than general theory, and being honest about the current state of things. The second file, findings.md, contains notes from a controlled test comparing three conditions: no instructions, an older version of the instructions, and the current version. The test was run on a real codebase to measure where these behavioral guidelines actually change an AI's output, where they have no effect, and whether the difference is worth the overhead of including them. The repository does not contain code. It is a reference document for people who work with AI coding tools and want to apply consistent behavioral guidelines across different models or projects. The README is sparse and points to the two markdown files for the substance. Licensed under MIT.
A small notes repo with a portable instruction set for AI coding assistants and a field test measuring whether those instructions change output.
Use freely for any purpose, including commercial use, as long as you keep the copyright notice.
Setup difficulty is rated easy, with roughly 5min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.