How abyakod/vibe-coding-harness works
This repository is a BKM reference for engineering AI-assisted software, illustrated by a URL-shortener demo. Its principal workflow is to define a spec, provide durable context, design validation before implementation, build from structured prompts, run the test harness, then request semantic review. The README documents that sequence and the demo’s capabilities; sampled code confirms the engineered shortener’s SQLite behavior and the judge’s review flow. The dashboard is a separate interactive, partly simulated presentation. Some capability links reflect documented guidance rather than inspected implementations.
Main entry points: SKILL.md Skill Manifest · app.py Engineered Shortener · llm_judge.py LLM Judge · context_spec.md Shortener Spec
Components 19
Engineering Guidance
- Skill Manifest
SKILL.md - Product Planning
skills/product_planning.md - Context Engineering
skills/context_engineering.md - Harness Engineering
skills/harness_engineering.md - Prompt Patterns
skills/prompt_patterns.md - Judge Guidance
skills/llm_judge.md - Tool Context Rules
.cursor/rules/harness.mdc - Copilot Instructions
.github/copilot-instructions.md
Shortener Demo
- Shortener Spec
demo_app/engineered/context_spec.md - Engineered Shortener
demo_app/engineered/app.py - SQLite Database
- Test Harness
demo_app/engineered/test_harness.py - Naive Shortener
demo_app/chaos/app.py
Semantic Review
- LLM Judge
demo_app/llm_judge.py - Anthropic API
Comparison Dashboard
- Dashboard Page
demo_app/web/index.html - Dashboard Logic
demo_app/web/app.js
Other components
- Engineer
- Dashboard Visitor
Connections 25
- Engineer → Skill Manifest: loads
- Skill Manifest → Product Planning: lists
- Skill Manifest → Context Engineering: lists
- Skill Manifest → Harness Engineering: lists
- Skill Manifest → Prompt Patterns: lists
- Skill Manifest → Judge Guidance: lists
- Product Planning → Shortener Spec: defines spec
- Context Engineering → Shortener Spec: provides context
- Harness Engineering → Test Harness: guides design
- Prompt Patterns → Engineered Shortener: guides implementation
- Shortener Spec → Engineered Shortener: specifies behavior
- Engineered Shortener → SQLite Database: reads and writes
- Test Harness → Engineered Shortener: validates
- Naive Shortener → Dashboard Logic: comparison content
- Engineered Shortener → Dashboard Logic: demo output
- Engineer → LLM Judge: runs review
- LLM Judge → Engineered Shortener: reads code
- LLM Judge → Shortener Spec: reads spec
- LLM Judge → Anthropic API: requests review
- Judge Guidance → LLM Judge: guides review
- Dashboard Visitor → Dashboard Page: opens
- Dashboard Page → Dashboard Logic: loads
- Dashboard Visitor → Dashboard Logic: interacts
- Tool Context Rules → Engineer: guides sessions
- Copilot Instructions → Engineer: guides sessions