docs: put first-screen evidence in PRs without repeated approval

[loopx-agent]

Routine first-viewport fixes within an authorized product task currently require a separate owner-confirmation round before commit or push. This repeats already supplied direction and stalls reviewable work. This proposal replaces that blanket confirmation with first-viewport evidence directly in the PR description.

• Inspect the actual viewport, embed its screenshot/browser capture in the description, identify the tested revision and data/build boundary, and provide the relevant before/after and narrow view or available preview link. Evidence only in chat, a comment or a Todo is insufficient.
• Reuse scoped direction for ordinary refinements and new commits. Refresh materially changed screenshots without asking for the same approval again. Ask when a consequential choice is missing, scope exceeds the request, or the user explicitly reserved presentation approval.
• Exact-head review, required validation, merge/install/release, paid execution and sensitive publication authority remain separate. Historical approvals are retained.

AGENTS owns the rule; design/frontend delivery and the existing PR template point to the same evidence obligation. Four English/Chinese RFC pairs replace repeated prospective confirmation requirements so older roadmap or workspace prose cannot reintroduce the delay. The bounded future-facing pass reconciles duplicate policy wording instead of adding another workflow or approval state.

Written by: Codex (OpenAI model agent). Basis: maintainer-requested first-screen workflow change, fully described above. UI impact: none; this changes author/review instructions, not the rendered product. No screenshot is applicable to this documentation-only proposal.

Validation: docs-governance smoke passed; staged bilingual RFC diff check and whitespace check passed; old prospective first-screen approval references were reconciled, with historical approval records preserved; relative links and the new AGENTS anchor were inspected; public-boundary scan clean. Runtime/browser tests are not applicable because no executable or UI source changes. This is a normal review proposal, not a merge or installation claim.

[loopx-agent]

Reviewer: model_agent; gpt-6.1-sol; OpenAI; runtime_reported; reasoning_effort=max

Approval conclusion (author-owned PR; GitHub blocks formal self-approval)

HEAD: a9f03c2dd464988aabcf1a0da07df0be68d47d8d

动机

受影响的是已收到明确产品任务和设计方向、准备提交首屏改动的 Agent 与审阅者。此前即使意图与权限已经给定，每次首屏或新提交仍可能追加一次确认，把本可直接审查的细化留在聊天等待。本提案把具体首屏、版本和数据/打包边界直接纳入 PR description；普通授权细化复用已有方向，只有真实缺失决定、越界或明确保留审批时才询问。这只改变拟采用的作者/审阅规则，不提交渲染界面，不授予 merge、安装、release、付费执行或敏感披露权限。剩余边界是文档提案尚未合并/安装，后续真实作者及模型对规则的采用未验证；历史批准仍按其原范围保留。

规格依据：AGENTS.md，不可变版本 fdf03ec6112aedf497b7ab1d6e5bc4db19c2ae5f。其中旧通用确认条款是本人要求本提案修订的对象，不把旧条款冒称为已满足的新规则。逐项：First-Screen Review Gate：Use current scoped direction to replace blanket first-screen re-confirmation with concrete PR-description evidence.（implemented）；Public And Private Boundary：Publish only public-safe visual evidence; authorized access alone is not publication authority.（implemented）；Author Submission And Review Entry：Exact-head review, validation and explicit merge authority stay separate.（implemented）；Installed adoption：Real installed author/model adoption is not established by this proposal.（deferred）。

改动思路

架构判断是复用 AGENTS 这个已有规则 owner，由 PR 模板及设计/交付/RFC 引用同一证据义务；无需再造审批状态或执行器。当前边界是首屏作者与审阅说明的完整修正，保留原精确头、验证、发布权限与隐私规则；不把文档规则视为已安装 Agent 行为。

条件	拟采用的行为	保留边界
意图、设计和权限已明确	首屏证据进入 PR，零追加确认轮	独立精确头评审和验证
可见结果实质变化	更新截图、版本与边界	不借旧图冒充新提交
真实缺失决定或明确保留审批	提出具体问题，继续独立工作	合并、安装、费用和披露单独授权

具体改动

12 条路径围绕一个已有政策 owner 同步：主规则说明实际首屏证据的提交时机、来源、版本和作用域；现有 PR 模板把证据字段放进 description；设计与前端交付指南引用同一锚点，四组中英文 RFC 更新未来门槛。已有权限、隐私、DCO 和精确头要求没有随之被弱化，也没有新增持久化状态、运行时开关或用户页面。

关键内容讲解

AGENTS.md 将通用的 commit/push 前额外确认改为提交/更新 PR 前查看实际首屏、在正文提供截图/浏览器图、before/after、桌面/窄屏与可用预览，标明 revision、数据和 prototype/packaged/installed 边界。实质变化须更新证据；明确保留的决定和越界仍询问，历史批准不扩张。

.github/PULL_REQUEST_TEMPLATE.md 复用 visual evidence 区并允许 UI 无变化时 N/A，继续禁止上传真实私有截图。design.md 与 frontend-delivery.md 使用同一新锚点。desktop-execution-frontends、goal-workspace-surface、live-team-workspace、overall-roadmap 的四组双语文件只替换未来重复确认/首屏门槛，不改原 V1–V12/L1–L3 功能验收、回滚和未交付状态；历史已获批准的叙述保留。

对主干的风险

没有发现本范围阻塞。独立 docs-governance、完整 committed bilingual diff、两条新锚点、公开扫描及 canary4 direct+11 selected 全通过，0 manual hold；唯一提交 DCO 正常。另逐字验证：删去被替换的首屏节并同步那处名称引用后，AGENTS 其他部分与基线完全相同。六个正反规则场景核对了已授权细化、新提交、真实缺失决定、明确保留审批、私有截图及独立 merge/install/paid 权限。

这是明确披露的作者/审阅默认行为变更，没有 opt-in，所以不冒称 feature-off 语义全等；变更范围外的准入及状态 owner 保持。证据要求是必须遵守的作者规则，并非机器自动批准或新执行凭证。未修改界面，无渲染截图适用，也未运行新模型来声称规则理解或安装采用。英文书内旧通用批准 checklist 由新根规则的优先条款覆盖，可在已有文档维护中同步，不作为新的审批门。

我的整体评价

APPROVE 这份完整规则修正提案。收益是让可审阅成果进入 PR，而把本人注意力留给真正缺失的决定；并保留所需证据和独立权限。前瞻整理已体现在同一 owner 的重复政策收敛，无需额外状态或框架。此结论不将提案当作已合并、已安装或证明模型长期遵循。

English verdict: APPROVE - HEAD a9f03c2; Scoped first-screen evidence rule reconciled across12paths, reserved decisions and effect authority preserved; docs/bilingual, clause counterfactuals and4direct+11selected checks pass. Installed/model adoption remains unverified.