fix: restore green main, and add the CI that would have caught it - #16
Merged
Conversation
`chore: remove unused files` deleted three icon assets and nine catalog entries but did not touch the code reading them, so main could not build or pass its tests. `go vet` stayed clean, which is why it went unnoticed: only the build and the test run reach these paths. The three assets are restored because agents.tsx never stopped importing them, and icons/agents.test.tsx asserts a distinct mark for cursor, openclaw and hermes -- it calls them PROMINENT_GUIDE_AGENTS, agents shown on the first screen alongside the one-click ones. That contract was not part of the cleanup. Those same three catalog entries come back, at ranks 6-8 rather than their old 3/5/6, since main renumbered the auto agents to 1-5 and the old values would now collide. Guide-only is not a dead concept: internal/app/install.go, catalog/manifest.go validation, AgentRow and ReviewPage all still branch on it, and provider_test.go and cmd/oneagent/main_test.go both use openclaw to exercise it. The six entries with no code referencing them -- kiro, gemini-cli, cline, continue, qwen-code, kilo-vscode -- stay deleted. The golden fixture is regenerated rather than hand-edited, so it comes from the same GetStatus call the assertion reads. It reorders more lines than it changes because Go marshals maps with sorted keys and the previous file was not written that way; the substance is six agents leaving and three staying. frontend/dist/.keep is added because it was never actually tracked. The .gitignore comment describes it as the thing that makes `go:embed all:frontend/dist` work on a fresh clone, and the negation on line 14 is right, but with no committed file the pattern had nothing to match -- so `go vet` on a clean checkout failed before any of the above could be seen. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The repository had no PR checks at all -- `gh pr checks` reported none, and build-artifacts.yml is workflow_dispatch-only and builds without testing. So the break this branch repairs reached main unopposed. Two jobs, matching what a developer runs locally: go vet plus go test -race, and the frontend test run plus build. The frontend build is the step that would have caught 4d63d75, since it resolves the imports that commit orphaned; `go vet` would not have, and did not. -race rather than a plain run because the write coordination lock in internal/app is not exercised otherwise. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The new CI run is the first time `go test` has run on Linux in this repository, and it failed on a test nobody had touched: TestTheRegionProbeRunsOnlyOnce, "region probe ran 5 times, want 1". The production code was right. countingRunner delegated to a real process.Runner, while the caller simulates darwin -- so the argv is `defaults read -g AppleLocale`. On a developer's Mac that succeeds and the answer is cached once, which is what the test asserts. On Linux `defaults` does not exist, so the probe reports "unanswered", and detectChineseRegion deliberately does not cache an unanswered probe: it retried five times, exactly as designed. countingRunner now answers the probe itself, so the test measures caching rather than which machine it runs on. Verified it still detects the regression it exists for: disabling the `if answered` cache in settings.go brings back "ran 5 times, want 1". Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Collaborator
Author
|
追加第三个修复：新 CI 自己抓到了一个我在本机看不到的问题。
生产代码是对的。 改成让 CI 现在两个 job 全绿——补 CI 的第一次运行就付出了价值。 |
This was referenced Aug 4, 2026
yujiezhang-ops
added a commit
that referenced
this pull request
Aug 4, 2026
…heir tools The repository had no LICENSE at all, which in a public repository means "all rights reserved": nobody could legally fork, modify or `go get` a module whose path is already github.com/MaimoryLab/OneAgent. Apache-2.0 was chosen for its explicit patent grant and trademark clause. NOTICE is the other half. distribution-compliance-policy.md has always required a third-party licence manifest as a release precondition, but the tool that generated it (cmd/oneagent-release) was removed in 23805b0, so the obligation had no implementation. Every entry is read from disk rather than assumed: the five Go modules actually linked into the desktop binary via `go list -deps`, the five npm packages in the bundle, and the two runtimes from runtimes.lock.json. The trademark section covers the eight Agent marks in frontend/src/components/icons/assets, which agents.tsx already documents as nominative use. Three documents are removed rather than edited. agent-installer-target-versions.md was internal strategy -- four unpublished conversion targets, SSO and unified billing plans, and a hosted model gateway that product-boundary-baseline.md explicitly lists as out of scope, so publishing both would have promised and denied the same thing. The two concept documents were generation logs: one carried a private image-proxy endpoint, and between them seventeen image links point into the gitignored output/ directory, so they would have rendered as broken images to any reader. Twelve documents cited cmd/oneagent-release, cmd/oneagent-rc or cmd/oneagent-provider-smoke, none of which exist. Removing them was deliberate -- 23805b0 replaced them with build-artifacts.yml -- so the docs are corrected rather than the tools restored. README and CLAUDE.md now describe the workflow that runs; historical documents keep their conclusions and gain a note saying the commands are background. Their command blocks are retagged from `bash` to `text`, because a header note does not stop someone scrolling to a shell block and running it. README also loses `task test:native`, a target no Taskfile defines. Two ADRs both claimed number 006. The public-site one becomes ADR-009, since ADR-008's Supersedes line refers to the multi-profile one; ADR-006 now says on its first line that its credential half is superseded, which previously appeared only sixty lines down. frontend/dist/.keep is restored: de3ed69 deleted it again, which re-broke `go vet` on a fresh clone in exactly the way #16 fixed. The cc-switch notes keep their findings but drop two conclusions that read as disparaging a named individual's project; both are rewritten as the methodological point they were actually making. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
修复 #15。
main当前在构建和测试上都是红的，本 PR 让它恢复绿色，并补上让这件事得以发生的缺口。根因
4d63d75（"chore: remove unused files"）删了文件，但没有动引用它们的代码：cursor.svg/hermes.png/openclaw.svg，但icons/agents.tsx:25,26,28的 import 还在 → 构建UNRESOLVED_IMPORT，7 个前端测试文件无法加载openclaw，golden fixture 也没更新 → 3 个 Go 测试失败go vet全程是干净的，这就是它没被发现的原因：只有构建和测试才会走到这些路径。二分确认过边界：
30d01f5（父提交）全绿，4d63d75三项全红。与 #4、#14 无关，我单独验证过在07cf902（#4 合并之前）就已经是坏的。恢复了什么，以及为什么不是全部
三个图标资源恢复，因为
agents.tsx从未停止 import 它们，而icons/agents.test.tsx:9明确把cursor、openclaw、hermes定义为PROMINENT_GUIDE_AGENTS——"和一键配置的 Agent 一起出现在第一屏，所以需要自己的图标"。这个契约不在那次清理的意图之内。同样这三个 catalog 条目恢复，rank 用 6/7/8 而不是原来的 3/5/6——main 已把 auto 条目重排为 1–5，沿用旧值会撞号。
guide-only 不是死概念，所以不能顺着删：
internal/app/install.go:340、catalog/manifest.go:129的校验、AgentRow.tsx、ReviewPage.tsx都在分支判断它，provider_test.go:79和cmd/oneagent/main_test.go:20更是直接用openclaw来测这条路径。另外 6 个条目保持删除：
kiro、gemini-cli、cline、continue、qwen-code、kilo-vscode——没有任何代码引用它们。这也意味着 #12 想提升的qwen-code仍然不在目录里，那个 PR 的决定权仍在你手上，本 PR 不替它做决定。golden fixture 是重新生成的，不是手改的
写了个临时 generator 走同一个
GetStatus调用产出 fixture（已在提交前删除），而不是从失败信息里抄。需要提醒的是这个 diff 比实质改动大：Go 序列化 map 时按键排序，而原文件不是那个顺序，所以重新生成必然带来重排。实质只有「6 个 Agent 离开、3 个留下」。我试过保留原顺序做最小 diff，但
catalog和supportedAgentIds是按 rank 排序的数组，顺序是语义而非格式，强行保留旧序反而会写进错误的排序——所以采用生成结果。frontend/dist/.keep这是 #15 里提到的第二个独立问题，一并修了。
.gitignore:14的!frontend/dist/.keep否定规则是正确的，注释说它是让go:embed all:frontend/dist在全新克隆上工作的关键——但那个文件从未被跟踪，所以规则没有东西可匹配。后果是全新克隆或新 worktree 上
go vet ./...会先失败于pattern all:frontend/dist: no matching files found，必须先跑一次前端构建才能编译 Go。验证方式是真的做了一次全新克隆：
.keep存在，go vet直接通过（此前不可能），go test和pnpm run build也都过。CI
仓库此前没有任何 PR 检查（
gh pr checks报 "no checks reported"），build-artifacts.yml是workflow_dispatch手动触发且只构建不测试。这就是断裂能畅通无阻进 main 的原因。新增
ci.yml，两个 job 对应本地实际跑的命令：go vet+go test -race，以及前端pnpm run test+pnpm run build。前端 build 正是能拦住4d63d75的那一步——它会解析那个提交遗弃的 import；go vet拦不住，事实也没拦住。用
-race而非普通运行，因为internal/app的写入协调锁在普通运行下不会被触发。验证
在本分支：
go vet ./...go test -race ./...pnpm run buildtsc --noEmit）pnpm run test外加全新克隆上从零验证：
go vet、go test、pnpm run build、pnpm run test151/151、覆盖率同上，全部通过。🤖 Generated with Claude Code