harness-golden-principles

Encode code review feedback as lint rules and run periodic scans generating small fix PRs.

Updated Jun 18, 2026
One-click install
npx skills add https://github.com/zhengguangli/harness-engineering-kit --skill harness-golden-principles-zhengguangli
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: harness-golden-principles
Source: https://github.com/zhengguangli/harness-engineering-kit/tree/main/skills/harness-golden-principles
Command: npx skills add https://github.com/zhengguangli/harness-engineering-kit --skill harness-golden-principles-zhengguangli

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? AI-generated code drifts into inconsistent patterns and repeated review feedback never becomes permanent enforcement, forcing teams to manually clean up code style over and over. ## Core Features & Use Cases - Golden Principle Distillation: Converts recurring review feedback and bug patterns into lint rules or documented taste principles, so human judgment is encoded once and enforced forever. - Periodic Entropy Sweeps: Delegates a read-only entropy-collector agent to scan the codebase on a fixed rhythm, classify deviations by risk, and produce fix PRs reviewable in under one minute. - Quality Tracking: Maintains quality score records and a tech-debt tracker so unprocessed findings are never silently discarded. - Use Case: Reviews repeatedly flag "don't use var, use const/let" — encode it as no-var and prefer-const lint rules, then let weekly scans drive violations to zero. ## Quick Start Ask the agent to distill the top recurring code review feedback into golden principles and set up a weekly scan that generates small auto-mergeable fix PRs.

Frequently Asked Questions about harness-golden-principles

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I turn repeated code review feedback into automated lint rules?▼

Distill recurring review comments into general principles, then check whether each can be written as a lint rule — if yes, encode it directly; if not, document it as a taste principle for periodic scanning. Each principle must trace back to real signals like review feedback or bug reports.

How to clean up inconsistent AI-generated code style automatically?▼

Run periodic entropy sweeps that compare the codebase against encoded golden principles using lint, grep, or semantic search. Each scan produces small fix PRs limited to one deviation type, with purely mechanical fixes tagged for auto-merge.

What is the difference between golden principles and architecture boundaries?▼

Golden principles govern taste and habit consistency through post-merge periodic sweeps, while architecture boundaries are structural invariants enforced as pre-merge CI blocks. Taste preferences must never become merge blockers, and structural constraints must never be treated as suggestions.

When should I not use periodic golden principle scanning?▼

Skip it when the project is extremely small with no repeated patterns, when structural architecture constraints are actually needed, or when the codebase is undergoing a major active refactor that would produce noise instead of signal.

How do I handle thousands of existing violations in a legacy codebase?▼

Use incremental adoption: scope each sweep to one module, file type, or pattern type rather than fixing everything at once. Record the full violation inventory in the tech-debt tracker and process it in prioritized batches.

Why should fix PRs be reviewable within one minute?▼

Oversized fix PRs create review backlog, get ignored, and break the sweep rhythm. Limiting each PR to one deviation type, at most 5 files and 50 lines, keeps reviews fast enough for the continuous cleanup cycle to sustain itself.