Repository navigation
Conversation
…llipsize, at desktop widths The 门禁 and 看板 columns of the settings project table are narrower than their state sentences, and the table's own white-space:nowrap (a ≥1280 rule) let the shared ellipsis cut「发布检查未完成」to「发布检…」. The spec drives #/settings/projects with one unconfigured gate and one collecting board per row, then reads both chips at 1280 and 1440: every rendered text line must lie inside the chip's own box on both axes (read from the line rects, because a wrapped flex chip can keep a ~2px scrollable extent while all its glyphs stay visible), the row keeps its 84px floor while growing with the wrapped chip, and the page does not scroll sideways. Below the 1280 breakpoint the card layout already fits every chip, and the spec holds it to that too.
… ellipsizing The settings project table scoped white-space:nowrap to its status chips at ≥1280, but its 门禁 (66–96px) and 看板 (next to the row switch) columns are narrower than the state sentences, so the shared ellipsis rule cut「发布检查未完成」to「发布检…」— the table now wraps: the chip folds to another line and the row grows with it. The base .status rule keeps nowrap for everywhere else, so #/projects chips (laid out in .project-features) and the below-1280 card layout are unchanged.
…er a one-character column The agent-CI-Infra#1011 review measured the 看板 switch — a 96px pill sharing the cell under the table's nowrap — parked on top of three of the folded chip's five characters: every line rect stayed inside the chip's box and this spec stayed green while the sentence was unreadable on screen. Readability is now read the way an administrator does: the topmost element at the centre and the quarter points of every rendered line must be the chip or one of its descendants (elementFromPoint over the already-computed line rects), and the 看板 chip must not fold into a one-character-per-line column (lines <= 2 for 看板采集中).
…instead of on the chip The 看板 cell of the settings project table also carries the row's board switch (static/views/project-list.js:64), and the table's ≥1280 nowrap kept both on one line in a column narrower than either: the chip folded into a one-character-per-line column while the 96px switch, unable to shrink below its own content, parked on top of it and covered three characters — agent-CI-Infra#1011's review measured elementFromPoint hitting the switch for 板/采/集 at both 1280 and 1440. This one cell now wraps instead: the chip takes the whole column on its first line and the switch sits below it, so neither paints over the other. The rule stays inside the @media(min-width:1280px) block that owns the nowrap it overrides, so the below-1280 labelled card layout is untouched and the 门禁 chip keeps the table's nowrap.
…ot counted twice, and assistant-homes gets its own row
…o other row counts, and assistant-homes leaves "other"
… not already counted in known The copy claimed every labelled volume's bytes were already inside known. Only bot-data, manager-data and the in-use *-work volumes are; agent-tools, the *-keys, *-notes and the *-work volumes left behind by deleted projects or older generations are excluded from the Docker row and have no known row of their own, so their bytes stay in other. Wording only; behaviour is unchanged. Also drops assistant_homes from the spec's unmeasured patch, since it is a measured row.
…rk volume, a deleted project included
…, in all three states
…under docker/
Move all four image Dockerfiles (control-plane, manager, project-computer, tunnel), the container Chromium wrapper script, and compose.yaml / compose.build.yaml into docker/, removing the single-file top-level images/ directory while keeping .dockerignore at the repository root.
Use context: ${BOT_BUILD_CONTEXT:-..} in docker/compose.build.yaml so both direct docker compose -f docker/compose.yaml -f docker/compose.build.yaml runs and --project-directory . local deployments resolve the repository root cleanly. Update workflows, acceptance stack, Docker/unit/e2e tests, public-board About link, and deployment/development docs accordingly.
Refs agent-CI-Infra#1051
…s from GitHub's facts The control plane gathers who is assigned, other people's open pull requests that GitHub links to the Issue (cross-references and closing references) and the latest comments as written. The Issue review reads them in its context and states in its recommendation whether someone else has claimed the Issue; a published review hands them to the Manager, whose built-in rule now asks that nobody else has claimed the Issue instead of that nobody is assigned. Nothing matches the text of a comment.
…skips a heading level
…ings at h2 The board hides the shared toolbar heading in board mode, so it carries its own sr-only h1. Settings cards and the identity overview can render their title as h2; the element name never sets the size, the card tokens style both levels.
The console-wide h2 rule would restyle a card title, so the h2 rules restate the h3 values; the rate-limit card's nested account name becomes h3 but keeps the h4 size and margin it rendered with.
…page, not four routes The blocking review was right: the spec walked the four audited list routes, so CI stayed green while #/settings/work-computers (the settings entry) and the three detail pages kept the h1 -> h3 jump. The categories now come from the settings registry the shell itself uses (settingsSections in static/settings/nav.js), and the detail pages come from the detail links the list cards render, so a section or detail page added later is covered without editing this file. Every route is judged by two rules: exactly one visible h1 first, and no level skipped between consecutive visible headings.
… h1, sub-sections are h3 card() keeps its h3 default for every page outside settings; a new settingsCard() wrapper (default h2) is used only by the settings modules, and liveCard()/settingsList() default to h2 because nothing outside static/settings/ calls them. Detail cards and the models project catalog promote their nested h4s to h3, and .card-sub restates the h4 size, weight and spacing, so the outline changes and the pixels do not. The visual rollout follows the .rate-account precedent. Also the two non-blocking notes: test/concurrency.test.mjs reads card titles in document order instead of all h2 then all h3, and test/console-tokens.test.mjs pins the new class/token rules.
…ettings card by its old h3 CI on 3c8b5f3 was red on exactly these: storage-retention-labels loads static/settings/page.js without static/settings/nav.js, so settingsCard() was undefined; project-list clicks the 项目管理 card heading to dismiss a menu; public-access lists the card titles. All three now read the h2 card title (or load the module that defines settingsCard).
…ttings outline, cover #/identity/feishu serviceRoleCard inherited the new settings liveCard default (h2), so the same card on #/identity/feishu became h3 -> h2 -> h4. The caller now names the level: settings passes h2, the Feishu page keeps h3, and the delete/save replacements follow it. e2e/heading-structure no longer waits a fixed 150 ms for the settings cards, it polls until the panel heading signature is non-empty and stable, so a slow page cannot be judged on the page title alone. e2e/channel-assistant-roles asserts the two regressions on #/identity/feishu (no h2 on the page, role card at its container level, 新建项目角色 one level below). Fixes agent-CI-Infra#1039
…, fail an unsettled outline serviceRoleCreate inserted the new card with the serviceRoleCard default h3 while every other card on the settings page is h2, so a created card read as a child of the card above it until a reload. The form now takes the same level as the card and settings passes 'h2'; the Feishu page keeps the h3 default. settledHeadings no longer returns its last sample when it runs out of attempts: exhaustion is a failed assertion, so a page whose headings never settle cannot pass the outline green. Refs agent-CI-Infra#1039
…eveloping label and one status comment While the bot develops an Issue the control plane keeps the project's developing label (new setting developing_label, default bot-developing, created in the repository when missing) on it and removes it once the pull request exists or the development stops without one. One status comment is created when the development starts and only edited afterwards: why it started, the plan the development agent reports with report_development_milestone (analysis, decisions and reasons, main changes, acceptance), and how it ended (the pull request with its key changes and tests, or why it stopped and what a person can do). It replaces the separate "Implemented in" comment on the Issue. Writes are kept in issue_development_progress; a failed write is retried by the engine's sweep with growing waits, which also settles developments that end outside a run. Onboarding plans at the develop level create the label.
…der test/ Move bot-acceptance/* to test/acceptance/*, real Docker system tests to test/docker/<module>/, opt-in live/probe scripts to test/live/<module>/, and test-runner orchestration files to test/framework/*. Upgrade test/framework/test-plan.mjs to discover test/ut/**/*.test.mjs and test/st/**/*.test.mjs recursively alongside flat test/*.test.mjs files while rejecting duplicate implicit test-set names across tiers. Upgrade scripts/check-runtime.mjs to syntax-check all JavaScript/ESM files recursively across runtime, static, scripts, public-board, test, and e2e directories in parallel. Refs agent-CI-Infra#1051
… development flow
Complete the final phase of agent-CI-Infra#1051 by organizing all remaining test assets under test/<category>/<module>/: - Split unit tests (test/ut/) and in-process service/integration tests (test/st/) into subdirectories mirroring the source tree (src/<module>/, manager/, runner/, static/, public-board/, docker/, workflows/, scripts/, framework/, acceptance/), and reject stray test/*.test.mjs files at the test/ root in test/framework/test-plan.mjs. - Consolidate Playwright configuration (test/e2e/playwright.config.mjs), browser runtime checks (test/e2e/browser/browser-runtime.mjs), shared browser helpers (test/e2e/helpers/), and all 196 browser specs under test/e2e/<module>/, removing the top-level e2e/ directory. - Move shared test helpers and fixtures into test/fixtures/ and move roadmap.live.test.mjs into test/live/tasks/. - Document the 8-tier test/ layout in docs/development.md and update file paths and Playwright config arguments across package.json, scripts/check-runtime.mjs, static/shell.css, and docs/**/*.md. Fixes agent-CI-Infra#1051
…dispatch system message A Manager on Pi (or any non-native backend) dispatched events with only its role prompt: runDispatchHarness got rolePrompt(role), so the dispatch prompt (the rule a project_rule basis must quote), the work the event can create and the write and merge rules never reached the model. Label-free development on the built-in rule was refused every time, and the merge rule was missing while the role could merge. The harness now gets the same system message as the native loop.
… codes and ids fold under 技术信息
… assistant panel agent-CI-Infra#1038 fixed the page view of a failed management open and left the floating panel out of scope; acceptance 3 asks both forms to behave the same. The card, the single statement of the reason, the two recovery links and the disabled header controls all come from the shared conversation surface, so this adds the coverage that pins it for the panel shape: the box on screen above the phone bar, every recovery control reachable inside the card's own scroll slot (never the page's), nothing able to send, and the panel never needing more scroll room than the page view at the same viewport.
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
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.
Written-by: LaunchPad Dev-4
Claim-Id: 6976637109
Refs agent-CI-Infra#616。本单验收第 3 条要求「浮动面板与全页行为一致」。agent-CI-Infra#1038 只覆盖了全页形态(其正文写明「浮动面板形态不在本次范围……另开处理」),这一支补上浮动面板形态的覆盖。
结论先说:浮动面板形态本来就已经一致,这一支不加生产代码
失败卡、单次原因、
重试/查看项目列表/查看身份与权限、新建会话/快捷操作的禁用与aria-describedby,全部出自同一处共享状态渲染(static/management/conversation.js:75-91的showAccessError)。浮动面板只是同一 DOM 的另一种呈现(.management-window:not(.management-page)),所以 agent-CI-Infra#1038 的修复在它身上同样生效。因此本 PR 只加覆盖用例,不为造红去改行为:此前这个形态没有任何 e2e 覆盖,
test/e2e/management/management-failure-state.spec.mjs的用例全部走#/management全页路由,浮动面板从没有被失败态覆盖过。加了什么
新文件
test/e2e/management/management-floating-failure-state.spec.mjs,3 条:#console-mobile-nav的 top;原因只出现一次、.management-status为空;三个恢复控件在卡片自己的滚动槽内可达且elementFromPoint命中自身;页面自身不滚动;管理消息与发送都隐藏;两个 header 控件 disabled 且带aria-describedby;390×844 与 1440 不需要内滚动。红先说明(按实际写,不造红)
这一支在 base
23c1c591上没有可红的断言:按「为造红去改行为」是被禁止的,所以改为实测记录。base 上同一套读数(现读 tip 构建,#/projects→ 底栏助手 → 无已启动项目的 fixture):[left, top, right, bottom]重试/查看项目列表/查看身份与权限)[10, 190, 380, 566]scrollTop=0即矩形与中心点都在槽内;第三条槽内滚 10px 后同样成立[10, 190, 380, 494][10, 190, 310, 494][10, 190, 380, 770][264, 260, 744, 1000][936, 136, 1416, 876]滚动只发生在卡片自己的槽里(
.management-access-state),页面scrollTop全程为 0;可达判据是「矩形在槽内 ±1px」且「中心点elementFromPoint命中控件自身」,不用toBeInViewport({ratio: 1})(390×640 会因 0.28px 的亚像素差报 0.9946)。内滚动需求对比(浮动 / 全页):390×640 = 31 / 56,390×568 = 103 / 128,320×568 = 166 / 201 —— 浮动面板每一档都比全页更省,这就是「一致」的可度量含义,也是第 3 条断言。
绿
npx playwright test -c test/e2e/playwright.config.mjs test/e2e/management/management-floating-failure-state.spec.mjs --workers=1→ 3 passed (8.9s)management-failure-state.spec.mjs同跑 → 5 passed (11.8s)--repeat-each=2→ 6 passed (17.3s)npm run check→Runtime JavaScript syntax checks passed (726 files)npm run build→✓ built in 1.30sUI 证据(before / after)
before 用 main
23c1c591构建,after 用本支构建,同一夹具、同一路由、fullPage: true,画面里没有 token / cookie / 审批链接。落盘尺寸按 PNG 头实测:floating-390-*= 390×876,floating-1440-*= 1440×900。fullPage390×876)fullPage1440×900)两对 before/after 的 sha256 完全相同(390 档
69d7f00072287c444029c2eeb0f6aab081067cc428bd04c8d0365b124d70ffd8,1440 档6581dc27859bda2c39c85789ff183d7bb08f7c5720cfc49537f91333ab3feffb):本支只加测试,渲染没有变,这是实测结论而不是省事。与 open PR 的文件自查
用连接器逐个读了当前 open PR 的改动文件(
#107114 files、#106711、#10613、#10032、#9574、#9157、#8883、#58212、#58115、#57719、#4903、#4437):没有一个改动test/e2e/management/或static/management*,与本切片零重叠。#581里的e2e/management.spec.mjs是目录迁移前的老路径(现树已无此文件),不算重叠。没做的
consoleState、static/app/states.js、static/settings/page.js、#481/#496导航。无 UI 面的声明必要——UI 证据照旧给了。