📡 Idea Radar 每天 8:00 / 18:00 推送 · 18:00 生成精选
reddit · SideProject · 55 分 ← 返回列表

I built Redpen to check what a coding agent actually finished, not just what it says

The test passes after the fix. But did it fail before the fix? That's one of the checks I'm building into Redpen, an open-source local CLI for the handoff after an agent says done. You define the task contract before coding. Redpen saves a Git baseline, imports completion claims from Codex, then independently runs applicable checks and produces an evidence receipt. Other agents can supply claims manually. It separates PROVEN, FAILED and UNVERIFIED instead of treating the final message as

五维评分

付费意愿
35
需求真实度
50
个人可落地
85
竞争空间
55
新鲜度
82

AI 分析

痛点:AI 编码 agent 声称任务已完成,但开发者无法信任其结论,缺少独立验证「是否真的做完了」的证据。

目标用户:使用 Codex 等编码 agent 的开发者、需要审核 AI 提交的团队 tech lead。

证据:I'm looking for developers willing to try one real bug fix and tell me whether the receipt makes review easier or just adds a step. / It separates PROVEN, FAILED and UNVERIFIED instead of treating the final message as proof.

变现:开源 CLI 免费引流,向团队版/CI 集成、云端证据回执托管、企业级 agent 验证合规报表收费。

MVP:2 周内做一个本地 CLI:保存 Git baseline、导入 agent 的完成声明、自动跑指定测试并生成 before/after 证据回执(PROVEN/FAILED/UNVERIFIED),先聚焦单一 JS node:test 场景。

竞品:CodeRabbit、Greptile、Cursor Bugbot、GitHub Copilot 内置检查

摘要:作者发布开源本地 CLI Redpen,在编码 agent 声称完成后自动保存 Git 基线、独立复跑测试并生成证据回执,区分 PROVEN/FAILED/UNVERIFIED。目前 before/after 校验仅支持单个无依赖的 JS node:test 用例,作者正在找开发者试用真实 bug 修复以验证回执是否简化 review。帖子为自荐,无明确付费意愿或收入数据。