它将可执行的纪律（而非建议）打包为可加载的规则，配有红绿验证、针对模型借口的合理化表，以及同步/漂移检测工具，确保跨会话和机器的一致性。
English | 繁體中文
A judgment layer for AI coding agents — make your AI think, verify, and communicate like a senior engineer.
MIT licensed · Works with Claude Code & Codex · Pure Markdown + two small Python scripts, no dependencies
Most AI agent failures are not knowledge failures. The model knows how to write the code. What it lacks is judgment — the operating discipline a senior engineer applies without thinking:
| What the agent does | What a senior engineer does |
|---|---|
| Edits a file, says "Done! Fixed the bug." — never ran anything | Runs the test, pastes the output, then says done |
| Bumps the timeout because you asked, knowing it won't help | Tells you the timeout isn't the cause, finds what is |
| "This should be much faster now" | "3.4s → 0.06s, measured" |
| Patches the symptom so the error message goes away | States the mechanism in one sentence, fixes that |
| Diagnoses the real bug, then asks "want me to fix it?" | Fixes it — it's in scope and reversible |
| Explains away a surprising test result to keep its story tidy | Stops and digs — surprise is where the real bug lives |
| Answers a one-line question with a 5-section report | Answers in a sentence |
Fable Soul packages the right-hand column as loadable rules. Every rule was captured from a real agent failure, written to prevent that specific failure, and tested to confirm it actually changes behavior. It is not a collection of "best practices" prose — it is operating infrastructure.
fable-soul/
├── SKILL.md # Skill entry point (3 modes: Embody / Maintain / Transfer)
├── references/
│ ├── soul.md # The judgment rules — the core of this project
│ ├── soul-compact.md # Token-lean rendering installed into global files (~40% smaller)
│ ├── maintenance.md # Mirror map, sync procedure, failure-capture loop
│ ├── transfer-prompts.md # 6 prompt templates for turning experience into skills
│ ├── evals.md # Behavioral eval scenarios + recorded RED-GREEN runs
│ └── worked-examples.md # Captured failures with before/after receipts
├── examples/
│ └── hooks.json # Optional Stop-hook enforcement for Claude Code
└── scri
A judgment layer for AI coding agents - make your AI think, verify, and communicate like a senior engineer
根据分类、Topic 和编程语言匹配的相似项目。
Harness Engineering is a methodology for improving coding agent outputs by carefully crafting the environment around them—providing curated context, tools, and executable constraints that encode an organization’s nonfunctional requirements and cumulative lessons.
A 28.9 million parameter language model runs on an $8 ESP32-S3 microcontroller entirely on-device, generating simple stories at about 9.5 tokens per second.
mimic captures traffic from any iOS or web app and automatically generates a Python client library that lets you call the app's API like a regular library.