Deep review of codegenie: architecture, specs, past plans and competing products

オープン
#44 コメント 6 件 リアクション 0 件 担当者 0 名 GitHub で見る

まだ誰も着手していません。

評価

難易度
5/5
見積もり時間
1週間以上
初心者へのやさしさ
15/100
issue の種類
ドキュメント
明瞭さ
明確に書かれている
活発さ
活発
技術スタック
typescript

調査の方向性

Start by mapping every top-level module under src/ against the specified specs, plans, tests, evals, and action configuration; record file and line evidence for each finding. Then review provider behavior, compare at least two fast lane designs, and research the named competitors with sources. Finish with four workstream notes and a ranked synthesis that includes measurements, risks, proposed tickets, and verdicts on #29, #30, and #31; the payload describes no specific test command.

索引モデルが issue の本文から書いたものです。

説明

Goal

A thorough, evidence-based review of codegenie as it stands at 0.8.0 (master), to harden its core goal: a high-fidelity, high-signal deep review harness that works across model families (Anthropic, OpenAI, OpenRouter and others), with skills layered on top. The review asks:

  1. Which building blocks can be simpler: merged, deleted or replaced by something already in the codebase?
  2. Where can it improve the signal (recall of material bugs, precision, calibrated confidence) and robustness across model families?
  3. Fast lane: a brainstorm and design options for a quick, cheaper review mode next to today's deep default.
  4. What competing or adjacent products do that is worth learning from (for example Sentry's Warden and Claude Code's ultra review), and what codegenie should deliberately not copy.

This is a research and review task. It changes no code.

Scope

  • All of src/, action.yml, bundled-skills/, evals/ and tests/, read against specs/project/, specs/plans/ (01–125) and specs/reviews/.
  • The open Backlog tickets #29, #30 and #31: is each still valid at 0.8.0, and how does it relate to the findings?
  • Web research on competing and adjacent review tools.

Ground rules

  • Code is the source of truth. Specs and plans are intent at the time they were written. Classify each divergence as stale spec, code bug, or a decision needed.
  • Prefer slim solutions. Deleting code counts as success. Any new machinery needs a measurement and a kill criterion. Reuse existing conventions, and add no new tunable thresholds.
  • Every finding cites evidence (file:line, a plan, a run artifact or a URL) and says how it would be measured.
  • No paid model runs (codegenie reviews, eval suites) without the operator's approval.

Deliverables

Notes in notes/research/, following the factory notes convention:

  • one note per workstream (pipeline architecture; models and providers plus the fast lane; code-level audit and tests/evals; plan history plus competitive research)
  • one synthesis note: ranked recommendations, each with evidence, expected impact, risk and a measurement or kill criterion, plus proposed follow-up tickets and a verdict on #29, #30 and #31.

Follow-up tickets go to Backlog only after the owner reviews the synthesis.

Owners

peter/@architect (synthesis), peter/@architect-2, peter/@lead-dev, peter/@lead-dev-2.

Acceptance criteria

  • Every top-level module under src/ is covered by at least one workstream note.
  • Every plan in specs/plans/ is accounted for in the history note, grouped by theme, with the outcome (still live, superseded, reverted, or never built) checked against the code.
  • The competitive note covers at least Sentry's Warden and Claude Code's ultra review, plus the other notable AI code review tools, with sources.
  • The fast lane gets at least two design options, compared on cost, latency and expected signal loss, with a recommendation.
  • Each workstream note was challenged by at least one other member, and the disagreements are resolved or recorded.
  • The synthesis is ranked and evidence-backed, and lists the proposed tickets.
主要言語
TypeScript
スター
8
フォーク
1
平均マージ
5分
マージ済み PR(30日)
2

環境構築

このプロジェクトには開発コンテナ、Dockerfile、コントリビューションガイドがありません。まず README を読み、一般的な手順ははじめてのコントリビューションガイドを参照してください。

はじめの一歩

  1. issue を最後まで読み、次にプロジェクトのコントリビューションガイドを読みます。
  2. 着手することを issue にコメントします — 二人が同じ作業をするのを防げます。
  3. リポジトリをフォークし、ブランチを切って変更します。
  4. issue 番号を参照したプルリクエストを送ります。

0xPolygon/codegenie のほかの issue

0xPolygon/codegenie の issue をすべて見る

似ている issue

TypeScript の issue をもっと見る

新しい issue をメールで受け取る

初心者向けの GitHub issue を短くまとめたダイジェスト。