Add debug-hunter — closed-loop FinTech debugging plugin for Claude Code & Codex

[ritim]

Project

debug-hunter — an open-source (MIT) FinTech AI closed-loop debugging framework, packaged as an installable plugin for both Claude Code and Codex CLI.

Proposed category

🛠️ Claude Code & Model Context Protocol (MCP) → 🤖 Claude Code (or a new "Community plugins" subsection if you prefer to keep that section official-only).

Proposed entry (per contributing.md format)

- [debug-hunter](https://github.com/yao-beyond/debug-hunter) - Closed-loop FinTech debugging plugin for Claude Code and Codex; detects financial correctness and security/fraud bugs with a knowledge base, Semgrep/CodeQL rules, and invariant checks.

Why it might fit

• Installable plugin on both platforms — /plugin marketplace add yao-beyond/debug-hunter (Claude Code) and codex plugin marketplace add yao-beyond/debug-hunter (Codex). Ships 7 subagents, a skill, and a /debug-hunter:debug-hunt command.
• Meaningful functionality — a 7-stage loop (threat-model → detect → reproduce-with-PoC → fix → verify-against-invariants → recycle) targeting financial bugs like IDOR over-withdrawal, TOCTOU double-spend, oracle manipulation, and rounding/precision errors. 30 documented patterns, 23 runnable pure-JDK demos, plus Semgrep + CodeQL rules.
• Documentation — bilingual README (正體中文 / English), CHANGELOG, knowledge-base map, tagged v2.0.0 release.
• Actively maintained — MIT licensed, recent commits, green CI (Semgrep 7/7, 23/23 demos, CodeQL).

Honest note on maturity

It's a young project (~3 weeks old, a handful of stars), so it may not yet clear the community-engagement bar in your guidelines. Sharing it for your consideration — entirely at the maintainers' discretion. Happy to open a PR instead if you'd prefer.