harness-engineering

Converts repeated coding-agent mistakes into repository instructions, checks, and failure memory.

25|4|Updated Dec 21, 2025
One-click install
npx skills add https://github.com/thangchung/agent-engineering-experiment --skill harness-engineering-thangchung
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: harness-engineering
Source: https://github.com/thangchung/agent-engineering-experiment/tree/main/agentgateway-entraid-obo/.claude/skills/harness-engineering
Command: npx skills add https://github.com/thangchung/agent-engineering-experiment --skill harness-engineering-thangchung

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Coding agents like GitHub Copilot repeat the same mistakes in a repository because project rules live only in reviewers' heads. This Skill turns those recurring failures into durable repository artifacts: agent instructions, enforceable checks, drift detection, and failure memory. ## Core Features & Use Cases - Repository Discovery: Inspects README, AGENTS.md, copilot-instructions, workflows, and package manifests before proposing any harness change. - Harness Adoption Workflow: Writes target-specific agent instructions, adds enforceable checks for high-value rules, records failure memory under docs/failures, and adds drift checks for guidance that can go stale. - Harness Review: Reviews existing or proposed harness changes from an opposing perspective, flagging generic rules, duplicate instructions, and unenforced high-risk rules. - Use Case: A team notices Copilot keeps editing generated API clients. Use this Skill to add a repository rule, a diff-scanning check that fails on generated paths, and a failure note documenting the incident and its prevention. ## Quick Start Ask the agent to adopt harness engineering for this repository so repeated coding-agent mistakes become durable instructions, checks, and failure memory.

Frequently Asked Questions about harness-engineering

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

FAQPage Schema
How do I stop GitHub Copilot from repeating the same mistakes in my repo?▼

Record each recurring failure as a durable repository artifact: a rule in AGENTS.md or .github/copilot-instructions.md, an enforceable check such as a script or CI gate, and a failure note under docs/failures naming the prevention. Harness engineering provides this exact adoption workflow.

What is harness engineering for coding agents?▼

Harness engineering is the practice of turning repeated coding-agent failures into repository artifacts: instructions, constraints, feedback checks, failure memory, evaluation, and governance. It treats the target repository as the source of truth and adapts every artifact to real evidence found there.

Where should agent instructions live in a GitHub repository?▼

Use AGENTS.md or .github/copilot-instructions.md for always-on behavior, and .github/instructions/*.instructions.md for file-scoped guidance. If the repository already has an equivalent location, update it instead of creating a parallel instruction system.

When should I not use harness engineering?▼

Do not use it for ordinary feature implementation; it applies only when improving the repository's agent operating environment. Also avoid it when automation would be brittle or misleading, in which case a documented manual review point is the better choice.

What is a drift check in a repository harness?▼

A drift check is a small script that detects guidance that has silently become stale, such as docs referencing removed commands or failure notes missing regression checks. It is typically written in the repository's existing language, with standard-library Python as a portable default.