adamentwistle/fable-skills — explained in plain English
Analysis updated 2026-05-18
Load engineering discipline skills into Claude Code so it follows habits like root-cause debugging and security checks by default.
Paste individual skill files into any AI tool's system prompt to guide its behavior on a specific task.
Review the included benchmark data to evaluate whether skill-guided AI output finds more bugs than unguided output.
Study the skill catalog as a checklist of best practices for careful software engineering.
| adamentwistle/fable-skills | angjustinl/adb-gopay | bbuf/kernel-pilot | |
|---|---|---|---|
| Stars | 89 | 89 | 90 |
| Language | Python | Python | Python |
| Setup difficulty | easy | hard | hard |
| Complexity | 2/5 | 4/5 | 5/5 |
| Audience | developer | developer | developer |
Figures from each repo's GitHub metadata at analysis time.
No dependencies, drop the skill folders into a Claude Code skills directory or paste them into a prompt.
fable-skills is a library of 35 small instruction files, called skills, meant to be loaded into an AI coding assistant like Claude Code. Each one encodes a habit that a careful, experienced engineer follows, such as reproducing a bug before fixing it, checking who is allowed to touch a piece of data before changing it, or sweeping through edge cases before calling a task done. The idea is that a strong model can follow these habits on its own sometimes, but loading the written checklist makes it happen on purpose every time instead of by chance. The skills are organized by topic: orientation and planning, correctness and debugging, security and safety, careful change discipline, and testing and verification. Each skill fires only when it is relevant. A debugging task pulls in the root cause debugging skill, a change touching user input pulls in the security reflexes skill, and a database schema change pulls in the data migration safety skill, so the model only carries in the guidance that actually applies to the task at hand. The repository includes benchmark results meant to show whether the skills help in practice. In one test, a security code review task was given to two fresh AI sessions, one with no guidance and one loaded with the relevant skills, then both reviews were scored blind by a separate AI grader. The version with skills found more of the planted issues, including an authorization flaw the unguided version missed entirely. A second, larger benchmark ran three different coding tasks five times each across two different AI models, and found that loading skills consistently raised scores for both models, and in some cases let a lighter model with skills match or beat a stronger model working without them. The README is upfront that these are small, directional benchmarks rather than statistical proof. Each skill lives in its own folder as a SKILL.md file with a short description and body. For Claude Code, you place the skill folders in a recognized skills location and the assistant picks the relevant one automatically. For any other AI tool, the files are plain Markdown that can be pasted directly into a prompt. The project has no external dependencies and is released under the MIT license.
fable-skills is a library of 35 loadable instruction files that teach AI coding assistants careful engineering habits like root-cause debugging, security checks, and edge-case sweeps.
Mainly Python. The stack also includes Python, Markdown, Claude Code.
MIT license: free to use, modify, and distribute, including commercially, with attribution.
Setup difficulty is rated easy, with roughly 5min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Don't trust strangers blindly. Verify against the repo.