Mango Plugins — AI skill for Claude Code

mango-plugins

  [](https://github.com/cuongdinhngo/mango-plugins/actions/workflows/validate.yml)

**mango** is a gated ticket workflow for Claude Code.

Instead of letting an AI coding agent decide a ticket is "done", mango forces every ticket through explicit review gates, produces evidence at each stage, and waits for your approval before continuing.

**Review evidence, not confidence.**

This repo is the **marketplace** that hosts mango ([`plugins/mango`](./plugins/mango)).

---

Why mango?

AI coding agents are excellent at writing code. They are much less reliable at deciding whether the work is actually complete. A ticket gets marked "done" while:

• one acceptance criterion was skipped
• a test proves the wrong thing
• an edge case was never considered
• documentation wasn't updated
• hidden assumptions remain

mango prevents this by routing every ticket through **gated phases** that each produce **checkable artifacts** before the next phase can begin. You approve a diff on evidence, not vibes. Silence is never approval.

---

What you get

Every ticket produces:

• ✅ A requirements matrix with explicit counts — nothing slips through unnamed
• ✅ A proving test at the right layer — a runtime requirement can't be closed by a unit-mock proof
• ✅ An independent challenger that never sees the work — it rebuilds the requirements from the raw ticket and flags anything the diff leaves unmet
• ✅ Human approval gates before every important transition
• ✅ A PR + tracker update + durable lesson, each behind explicit per-action approval

---

Quick start

In Claude Code:

/plugin marketplace add cuongdinhngo/mango-plugins
/plugin install mango@mango-plugins

/mango:init          # detect your stack, write .harness.json,