Skip to content

docs: add bilingual context delivery guide and qualification plan - #7806

Merged
huangruiteng merged 4 commits into
mainfrom
codex/context-delivery-guide-20261101
Nov 2, 2026
Merged

huangruiteng merged 4 commits into
mainfrom
codex/context-delivery-guide-20261101

Conversation

@loopx-agent

@loopx-agent loopx-agent commented Nov 1, 2026 •

Copy link
Copy Markdown
Collaborator

Goal And Delivered Outcome

上下文交付已有 reference、实现和分散的 RFC,但缺少一份从实际任务出发的完整使用与验收指南。新增中英文指南,贯通来源权威、capture、必要/支持材料、渐进阅读、同响应引用、精确差分、session/epoch、恢复和原 Turn 一次结算;从指南目录和 reference 接入。

同一 PR 更新成对 roadmap/RFC:文档先交付,再以 DeepSeek 与强模型分别做可比成果/完整成本对照,随后验收可回退 profile 采用/发布、默认范围扩展和实际旧路径退役。保留单会话原 quota 对照、历史最佳原生排名、失败/负例与未支持范围,不把较短输出写成已验证收益。

  • Outcome basis: 独立可阅读、可操作、可评审的上下文交付说明和最小双模型验收方案。
  • Observable before → after: 目录没有独立指南 → 双语入口、完整任务示例和下述实际 GitHub 渲染;静态检查覆盖新增链接。
  • Intended base: main,工作基线 111519af76e56141c56eb7a27443da59def775b9;不关闭整项 C1/C2/C9。

Author Declaration

  • Written by: model_agent — OpenAI Codex, GPT-6 family.

Implemented against

  • Specification and revision: accepted documentation request stated above; current semantics in docs/reference/required-work-context.md and the single-session validation scope in docs/architecture/rfcs/controller-executor-unified-contract-v0.md §9 at base 111519af76e56141c56eb7a27443da59def775b9.
Criterion Disposition Path Evidence
一份指南解释完整交付链,区分缓存/摘要/精确差分、当前实现/提案 implemented docs/guides/context-delivery{,.zh-CN}.md 逐段对照 current reference、TypeScript composition/codec 与 native adapter
同一任务的冷、暖、约束变化、压缩/中断恢复;设置、关闭、排障和隐私 implemented 两版指南 UTF-8 导出合成任务、支持矩阵、原 capture/原 effect 恢复及默认优先级
双语路线图与阶段顺序 implemented 成对 overall roadmap 和 controller/executor RFC staged、exact-head bilingual diff checks
最小 DeepSeek/强模型对照;固定条件、完整成本、失败及适用原生成果 implemented 指南最小研究与 RFC §9 新小节 四格对照、board/integrity 复用、未知不记零、单会话先行;仅交付方案
默认采用与旧路径退役基于实际证据 implemented 指南/roadmap/RFC 保留完整快照、摘要、权限/身份/停止、历史 reader 和回退;无生产行为变更
读者入口、链接、公开边界和实际渲染 implemented guides index、required-work-context、两版指南 36 个目标/锚点通过、8 个 Markdown 公开扫描通过、下方截图
  • Self-check: 已读双语指南与成对 diff,并核对当前 reference、真实 CLI 默认/off 路径和实际 GitHub 渲染。未运行模型或科学实验,未查询、等待或手动触发 CI;这些检查不代替模型采用、净费用或长程成果资格。

Scope And Continuation

  • Completed scope: 文档与验收方案完整,供普通 PR 评审。既有设置、执行入口、预算、时钟、评分和安装状态均未修改。
  • Remaining work / existing owners: C1 调用方、统一预算与真实最终输入资格,以及 C2/C9 的双模型证据、受支持 host/profile 采用和实际旧路径退出,继续由现有 runtime/host/benchmark/caller owner 完成。Codex adapter 不资格化 DeepSeek host;模型版本、价格和阈值未知时执行前登记,不编造、不以此阻塞文档交付。
  • Boundary: 独立文档增量可单独评审/回退;没有新实验、默认晋级、合并或发布授权。无需为本指南另建跟进任务。

Review Fix

修复原 P2:指南开篇与设置说明不再把“未配置”一概写成 off。受支持 POSIX Codex App/SSH 与已声明的 benchmark heartbeat,在机器偏好缺省时默认 on;saved/explicit off 保留,其他入口继承 full。显式开启的错误与不支持场景仍按 reference 处理。

成对指南、RFC 和 roadmap 同时区分当前有限默认与未来扩大默认范围的双模型资格。文档纠错不改变运行时默认,也不宣称现有默认已经证明模型收益。仍使用原 PR,正常快进更新到 7af1045410f19a8088a7da7b06c78ad339469e85。

Validation

  • Tested revision: 7af1045410f19a8088a7da7b06c78ad339469e85.
  • Verified base: main at 111519af76e56141c56eb7a27443da59def775b9; committed diff contains only the eight documentation files.
  • Run state: finished. Input classes: synthetic, public_fixture. Python 3.14.8, POSIX.
Check Result Scope and limits
Existing CLI/default/off tests 24 passed Run at 29b20a2; final layout-only commit has identical runtime/tests. Three native scenarios default on → saved off and real lifecycle hook; ordinary CLI/full, worker off, other hosts, unsupported activation, File/SQLite sessionless same-capture recovery. Isolated fixtures; no live Goal mutation or model launch.
Standard premerge 4 direct + 15 selected passed uv run --extra test loopx canary premerge --from-git-diff --git-diff-base origin/main; includes exact-head bilingual gate, semantic vocabulary, maintainability, selected workflow smokes and public boundary. First attempt lacked npm development dependencies; after locked npm ci --ignore-scripts, the complete gate passed. Original failed attempt retained.
Links and public boundary 36 targets/anchors; 8 Markdown files passed Both new guides plus added links in the other six files; zero public-scan errors/warnings.
DCO and hygiene passed Original signed commits retained; signed merge/update commit, clean diff and worktree.
Actual rendered guides observed English/Chinese, 1280px and 390px; before/after evidence below. No document-wide horizontal overflow; activation choices reflow as five readable items.
Model effectiveness, billing, packaged/installed adoption not_run No model study or product installation. Native engineering checks do not qualify C1/C2/C9 or DeepSeek support.

Independent exact-head review, current change-quality/readiness and merge authorization remain separate. This author validation is not an approval or permission to merge.

Frontend / Visual Evidence

  • UI impact: documentation content only; no App or documentation chrome change.
  • Boundary: actual GitHub-rendered public Markdown in Ego Lite, reading viewports 1280×900 / 390×900. The guide's first heading/default behavior and changed activation section are captured; unrelated account navigation is outside the capture.
  • Before: 0b0c8b4f1796ab19e9077526d7b373a54f74e989; after: 7af1045410f19a8088a7da7b06c78ad339469e85.
  • Preview: English guide · 中文指南.
Desktop Before After
English English before English after
中文 中文 before 中文 after
390px first reading viewports and activation choices
First viewport Before After
English English narrow before English narrow after
中文 中文 narrow before 中文 narrow after
Activation choices Before After
English English choices before English choices after
中文 中文 choices before 中文 choices after

Attention review: 定义、当前默认边界和理解/使用/验收入口保持同一阅读列;必要开关后果与“保存开关不证明收益”分段。英文窄屏导航延伸到下一屏,正文完整可达,没有缩小字体来强塞首屏。实际检查发现,开关表的长 CLI 标记把效果列挤得很窄,已改为五条选择/后果列表:两版均保留显式选择、机器偏好、未配置、worker off 与 App off 的全部后果,390px 文档无横向溢出,五项均可按自然行宽阅读。其他对照表保留其原有局部滚动。GitHub 自有 chrome、字体和控件未改。首屏图中的 Mermaid 可能仍显示加载指示,未据此声明图形渲染通过;编号流程保持可读。

Technical Direction And Boundary

  • Existing C1/C2/C9 and the overall roadmap own caller migration, actual final-input/model outcomes, complete cost, reversible profile adoption and retirement. No parallel workflow or new experiment authority.
  • The guide/reference and paired RFC owners are reused; no new runtime owner, configuration, abstraction or smoke was introduced for this correction.
  • All changed files, description and attachments contain public/synthetic material. No private Goal state, credentials, local paths, raw benchmark/verifier evidence or internal links are published.
  • This is a completed documentation proposal for independent review. It does not close the broader runtime/research work or merge itself.

Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>
Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>
huangruiteng added a commit that referenced this pull request Nov 1, 2026
Signed-off-by: huangruiteng <huangrt01@163.com>
huangruiteng added a commit that referenced this pull request Nov 1, 2026
Signed-off-by: huangruiteng <huangrt01@163.com>

@loopx-agent loopx-agent left a comment

Copy link
Copy Markdown
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Reviewer: model_agent; gpt-6.1-sol; OpenAI; runtime_reported; reasoning effort: max

Request changes conclusion (author-owned PR; GitHub blocks formal self-review)

一项 P2 阻塞:新指南的缺省开关说明落后于当前目标主干。

动机 / Motivation

受影响的是通过指南理解、启用或关闭原生上下文交付的用户。用户在受支持的 Codex 原生入口上不保存机器偏好,指南告诉他交付关闭,当前主干实际启用;显式保存 off 才能可靠保持关闭。双语使用说明与研究方案已经可读,但默认值表和开篇未反映当前主干,暂不能作为可靠的现行操作指南交付。本次只交付双语指南和资格方案,不执行模型实验、不修改 runtime 默认值、不安装 Host,也不证明模型收益或整条路线已完成。

改动思路 / Approach

指南复用现有 reference、机器设置和研究 owner,不另建配置;但面向当前用户的默认行为必须与已合入实现一致,历史 base 上自洽不足以证明合入后正确。当前 PR 是八个文档文件的新增/导航及双语资格说明,修复范围应限于现行默认语义和对应使用说明。已从实际入口核对正向与拒绝路径;原有配置、决策与持久化权责继续由既有 owner 承担。不把保存设置、宿主准备、实际输入格式或模型收益混为一个验收;未来研究方案与现行行为分别核对。

具体改动 / Changes

审查 head 0b0c8b4f1796ab19e9077526d7b373a54f74e989,branch merge base 70f643153ea773b7bc1ef71bdb06095b0744bd6d。### 关键内容讲解

新增英文283行、中文225行指南,另六个文件加入导航或双语资格说明,共 +578/-0。两版都区分 progressive reading、精确 delta 与 provider cache;缺失证据回同一 capture、selected EOF 不等于全文已读、压缩不取消已准入 Turn、已发生效果先查原 receipt,这些边界与既有 owner 一致。研究方案保留单会话 original quota 对 composed path 的优先对照、DeepSeek 与强模型分开比较、真实最终输入/补读/恢复/完整成本,以及未知值和失败不抹去;它未授予实验执行权。

当前规格 docs/reference/required-work-context.md 固定于 73c5daab342afe8aa6c1a85dc92bc0bf352d9132:Selected capture details / 按需读取捕获详情 已落实;Machine default not_met。现行 reference 明确 supported POSIX Codex App/SSH/declared benchmark heartbeat 在偏好缺省时 on;普通 CLI、其他宿主、不支持平台保持 full,保存或显式 off 继续关闭。

[P2] 更新两版的缺省选择说明。 英文153行 写 absence resolves to off,中文124行 写缺省为 off;开篇也重复默认关闭。它们在旧 branch base 上自洽,但当前 主干 resolver 和上述 reference 已不同。请同步修正四处,明确保存 off 的关闭方式,并让未来资格要求与已发布的有限 native 默认范围相容;不要求回滚 runtime 或先执行模型实验。

对主干的风险 / Risks

PASS:4项直接和15项选中 premerge 检查、双语 RFC diff、所有提交 DCO、36个新增相对文件/锚点链接,以及8个文件的公开边界扫描。最初未指定 registry/runtime 的单独 public check 因本机双默认路由冲突失败;保留该次失败后改用显式空隔离路由,扫描0错误/0警告,未更改任何活跃配置。

FAIL:现行默认值一致性。独立切到固定主干后,真实 test_native_configuration_defaults_on_and_later_explicit_off 三个 supported-host 参数均通过,证明缺省 on、随后保存 off 的行为;指南的 off 声明因此确有冲突。静态全绿和旧 head 的说明一致不能替代当前目标事实。错误会影响新用户对启用状态的判断,故请求修改;只需修正文档并对当前 target 复核,不扩大到模型实验或默认值实现。

已实际检查 GitHub 双语首屏、目录前后、390px窄屏正文/表格、英文竖向 Mermaid 和中文资格研究后段。英文宽表可横向查看右侧义务,中文场景表直接适配;这些视觉证据不证明模型采用或收益。未查询或等待 CI。

我的整体评价 / Verdict

REQUEST_CHANGES:研究方案和恢复叙述是有用的文档成果,但现行默认值错误尚未修复,不能宣布完整指南交付。long_horizon 为 preserved:没有执行新循环、改写原任务或隐藏失败;user_experience 为 regression,原因是错误的开关状态预期,而不是 Markdown 排版。双语正文、分层入口、长页后段和窄屏表格已实际查看;可读性良好,但默认关闭的错误说明会误导用户判断当前状态,属于操作语义回归。未验证站外文档站、安装版 Host 或真实模型采用;中文 GitHub Mermaid 在捕获时仍显示加载,已检查同页完整编号正文,未把中文图计为渲染通过。请先修正 P2,在 unchanged new head 上复核 default/off 矩阵和双语链接/渲染;不因审查建议新建配置、批准流程或实验任务。

English verdict: REQUEST_CHANGES 0b0c8b4; both guides incorrectly describe absent machine preference as off, while current main enables supported native Codex scenarios. Update current-default guidance in both languages; static checks and the proposed study do not resolve this mismatch.

Comment thread docs/guides/context-delivery.md Resolved
Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>
Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>

@loopx-agent loopx-agent left a comment

Copy link
Copy Markdown
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

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

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

精确 head 7af1045410f19a8088a7da7b06c78ad339469e85 可以批准;原 P2 的双语默认值错误已独立核实修复。CI 未作为批准条件,当前 change-quality receipt 和合并授权仍是独立限制。

动机 / Motivation

需要理解或操作上下文交付、判断当前是否启用的用户与 Agent。例如机器尚未保存偏好,用户按旧稿以为原生路径关闭,但受支持的 POSIX Codex App/SSH/已声明 benchmark 心跳实际上默认开启;新版两种语言都明确缺省 on、保存或显式 off 可关闭,其他入口保持 full,并把未来扩大默认范围的资格与当前有限默认区分开。读者可以从指南首屏进入场景示例、启用/验证/回退和资格说明,按既有 reference 核对真实调用,而不另建配置。本次是完整的双语操作与理解文档,不执行模型实验,不证明模型已采用输入、长期成果提高、完整成本下降或旧路径可以退役。

改动思路 / Approach

复用 当前 Machine default 和既有设置/capture/回执 owner,把理解、使用、恢复与研究验收连成阅读路线。当前有限 native 默认与未来更广默认采用、旧路径退役分别判断,既不另建开关,也不要求为修正文档先跑模型实验。

正向路径从现有导航进入指南,再从四种场景转到启用、实际调用核验和原任务结算。负向路径中,基线丢失或引用失效应回同一当前 capture、必要时完整交付;保存 on、某次 full/delta、缓存命中、selected EOF 和模型理解不能互相替代。图只是解释,完整编号正文仍可独立阅读。

具体改动 / Changes

关键内容讲解

相对当前 target 111519af76e56141c56eb7a27443da59def775b9 的实际范围是八个文档文件 +595/-0,两版指南分别 292/228 行。早先 GitHub PR 索引仍以旧 base 展示 296 文件;作者刷新 base 后,重新取得的原生 packet、Git merge base、GitHub files 与比较接口均已确认这八个文件,没有把旧 296 文件范围当作已审完。

两篇英文/中文指南区分 progressive reading、exact delta 与 provider cache,保留来源、必要事实、权限、当前恢复和原 once-only settlement。开篇和五项设置选择均明确:支持的 POSIX native 调用在偏好缺省时 on;saved/explicit off 与 worker off 保留;普通 CLI、其他宿主及不支持平台保持 full。原线程中的 P2 及作者回复均已完整核对,讨论保持原状。

其余六个文件把既有导航接到指南,并在成对 RFC/roadmap 中说明未来资格。对应规格条款:Machine default → implemented,28 项独立现有默认/关闭/宿主/CLI 测试通过;Selected capture details → implemented,当前来源恢复、原效果和完整读取边界保留;C1 single-session; C2/C9 future qualification → implemented 为研究方案,保留同一会话原 quota 与 composed path 的首要对照、双模型分别固定版本/条件、最终输入/补读/恢复/成果/完整成本与失败。四格表不代表实验已执行,更不证明模型支持、收益或可退役。

对主干的风险 / Risks

最强风险是静态检查通过却给出错误操作预期,或把资格设计说成已证实收益。这次对照的是当前 target 的 reference/实现,且真实默认矩阵在本 head 通过;不是只在旧 branch base 上自洽。原 runtime、设置、权限和预算没有修改,不增加强制确认轮次;长期采用和成本证据仍未知。

PASS:28 项现有真实 CLI/default/off 测试(69.21 秒,零失败、零跳过)、36 个文件/锚点链接、四个提交 DCO,以及标准 premerge 的 4 direct +15 selected 命令检查。整体 premerge 仍 FAIL(exit 1),仅为 quality_stale_receipt,merge_gate_passed=false,零命令失败;未改写 receipt、未豁免合并门。CI 未查询或等待,作者历史失败也没有因此改记为通过。

我的整体评价 / Verdict

该双语指南的有用交付已完成,不必为文档批准强制要求未知 CI 转绿。long_horizon=preserved:来源、原效果、失败和研究资格界限保留;user_experience=improved:正确的默认预期、三条任务入口及连续的五项激活列表降低理解和恢复成本。相邻改进采用了保留全部语义的窄屏列表,无需新框架或 runtime 重构。

已实际查看 GitHub 精确 head 的双语 1280×900、390×900 首读正文,以及窄屏五项激活列表和中文长页资格表。标题、正文和三条任务入口层次清楚;五种选择及后果在窄屏连续可读,后段仍保留未知、失败与研究边界。未验证独立文档站、安装版 Host 或真实模型采用;中文 Mermaid 在首屏捕获时仍为加载状态,完整编号正文可读,未把中文图计为渲染通过。英文图已在实际截图中渲染。这些观察不证明模型采用、理解或长期净收益;C1 完整迁移和双模型后续验收仍未完成。原 unresolved thread、change-quality qualification、独立 readiness 和合并授权应继续由对应 owner 处理;本轮不 resolve、dismiss 或 merge。

English verdict: APPROVE 7af1045; the original bilingual current-default mismatch is fixed and independently covered by 28 local tests and rendered guide inspection. CI is not an approval prerequisite; quality qualification, unresolved discussion and merge authorization remain separate.

@huangruiteng
huangruiteng merged commit d724a4c into main Nov 2, 2026
28 of 40 checks passed
@huangruiteng
huangruiteng deleted the codex/context-delivery-guide-20261101 branch November 2, 2026 07:33
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants