git:20260311.7f217db to git:20260416.dd9f369

24 added, 23 removed. Audit A to A.

- # claude-review-loop

+ # harness-engineering-skills

- > Cross-LLM iterative code review plugin for Claude Code.

+ > Stometa's public curated Claude Code skillset — a small, opinionated set of engineering skills published as one Claude Code plugin, extracted periodically from the private `stometa-skillset`.

- ## What it does

+ ## What this repo is

- This Claude Code plugin spawns a peer reviewer (OpenAI Codex CLI or Google Gemini CLI) to independently review your code changes. Claude Code evaluates the peer's findings, implements accepted fixes, and re-submits for re-review. The loop iterates until both AI agents reach consensus on code quality.

+ This repository is the public publication surface for selected skills from Stometa's private `stometa-skillset`. The first batch ships a single Claude Code plugin, `harness-engineering-skills`, containing two skills: `review-loop` and `harness`. Future batches will add more skills as they stabilize internally.

- ## Key features

+ ## Skills in this release

- - Fully autonomous — no human participation needed during the loop

- - Cross-LLM diversity — catches blind spots that single-model review misses

- - Iterative convergence — findings are debated, not just listed

- - Structured output — JSON logs + markdown summary per session

- - Auto-detects review scope — local diff, branch commits, PR, or specific commit

+ - **review-loop** — Cross-LLM iterative code review. Spawns a peer reviewer (Codex CLI or Gemini CLI), evaluates its findings, implements accepted fixes, and re-submits until both sides converge. Runs autonomously; humans watch via `.review-loop/<session>/summary.md`. Install: `claude plugin install harness-engineering-skills@stometa` then invoke `/review-loop` inside Claude Code.

+ - **harness** — Cybernetics-based multi-agent orchestration for complex tasks. Runs a Planner → Generator → Evaluator → Retro pipeline with fresh sub-agents per checkpoint to prevent drift, and persistent retro learning across tasks. Recommended flow: Claude Code plans the spec (Session 1); Codex executes autonomously (Session 2); Claude CLI acts as the cross-model reviewer. Install: same plugin as above; invoke `harness plan <task-id>`.

- ## Installation

+ ## Install

```

- claude plugin marketplace add stone16/claude-review-loop

- claude plugin install stometa@claude-review-loop

+ claude plugin marketplace add https://github.com/stone16/harness-engineering-skills

+ claude plugin install harness-engineering-skills@stometa

```

- ## Usage

+ ## Prerequisites

- ```

- /stometa:review-loop

- ```

+ - Required: `git`, `python3`, Claude Code with the `superpowers` plugin.

+ - Peer reviewer (one of): `codex` CLI or `gemini` CLI — only needed when using `review-loop` or `harness`'s cross-model review gate.

+ - Optional: `gh` CLI for PR-scoped review detection.

- ## Prerequisites

+ ## Positioning notes

- - Claude Code CLI

- - Codex CLI (`npm i -g @openai/codex`) or Gemini CLI

- - Git

+ - This repo is not "claude-review-loop" anymore; `review-loop` is now one skill of several. The repo name is `harness-engineering-skills`.

+ - The marketplace name is `stometa`; the plugin name is `harness-engineering-skills`; together they become `harness-engineering-skills@stometa`.

+ - License: Apache-2.0.

- ## License

+ ## Links

- Apache-2.0

+ - [README (English)](README.md)

+ - [README (中文)](README.zh-CN.md)

+ - [LICENSE](LICENSE)

+ - GitHub: https://github.com/stone16/harness-engineering-skills

+ - Private origin (for context, not public): Stometa's `stometa-skillset`.