Horace-Maxwell

Horace-Maxwell / Harness_Engineering_Regression_Copilot

Public

Failure-first AI regression testing CLI for turning AI failures into local regression assets and PR gates. 把真实 AI 失败快速变成可执行回归资产和防止再次犯错清单。

11
1
100% credibility
Found Apr 11, 2026 at 11 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
JavaScript
AI Summary

A local tool that transforms real-world AI failures into simple, runnable tests for ongoing quality checks in development workflows.

How It Works

1
🕵️ Hear about HERC

You learn about this handy helper when your AI app starts making repeated mistakes in real use.

2
🏠 Set it up in your folder

You add it to your project's home so it's ready to catch errors right where your work lives.

3
📋 Paste in a failure

You copy a bad conversation or error example, and it saves it safely for review.

4
🔍 Turn mistake into test

It smartly shapes the failure into a reusable check to spot the same problem again.

5
✅ Review and save good answer

You check the test details and add what a correct response should look like.

6
🚀 Run safety checks

You press go and instantly see if your AI behaves correctly on known trouble spots.

📊 Get clear reports

You receive simple summaries showing what's fixed, preventing old errors from sneaking back in.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 11 to 11 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is Harness_Engineering_Regression_Copilot?

Harness_Engineering_Regression_Copilot (HERC) is a JavaScript CLI for failure-first AI regression testing in AI apps, RAG systems, agents, and copilots. It turns real-world failures—like bad conversations, broken traces, or support tickets—into local regression assets, baseline responses, and PR gates with commands like `herc import`, `distill`, `run`, and `report`. Developers get a repo-local loop to capture issues fast and block them deterministically in CI, without cloud evals.

Why is it gaining traction?

Its local-first, deterministic gates run in 464ms for first fails and scale to 691 cases/sec, slashing CI from 5000 to 3 changed-only cases. Benchmarks show 100% failure leakage reduction and shipped correctness from 92.8% to 100% on 920 instructions. The CLI hooks into GitHub PRs via exit codes, JSON reports, and `--changed`, making engineering regression feel lightweight at 55KB packed.

Who should use this?

AI engineers triaging production failures in chatbots or RAG pipelines. Support and QA teams converting tickets into executable tests. Platform devs gating PRs for agent workflows or policy copilots, especially if you're tired of manual repros.

Verdict

Try it if you're building AI with recurring failures—solid docs, repro benchmarks, and cross-platform CI make onboarding easy despite 11 stars and 1.0% credibility score. Still early; watch for adoption before production gates.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.