Skip to content

Keep ready controls visible for large Todo scopes - #9106

Open
loopx-agent wants to merge 1 commit into
mainfrom
codex/ready-todo-scope-navigation-20261120
Open

loopx-agent wants to merge 1 commit into
mainfrom
codex/ready-todo-scope-navigation-20261120

Conversation

@loopx-agent

@loopx-agent loopx-agent commented Nov 20, 2026 •

Copy link
Copy Markdown
Collaborator

Goal And Delivered Outcome

A ready native quota decision can carry a long Goal write scope and a long list of requested Todo scopes. The bounded reader already deferred the Goal list, but the Todo list could still make the opening too large. The first page then ended inside the Goal scope before the stop, required reads and readiness were available.

The ready reader now presents those controls and the selected Todo text in a complete opening at the existing page budget. Its linked read supplies the two exact scope lists and any deferred approval entries from the original capture. Previously observed Goal scope remains eligible for reuse. Losing the observation restores the complete linked read.

Author Declaration

Written by model_agent (OpenAI Codex, GPT-6).

Implemented against

The native ready read and recovery clauses in the required work context contract.

Criterion Disposition Evidence
Deliver the current decision, stop and required reads in one bounded opening when the controls fit. Implemented Both providers expose the complete opening with 290 Goal scopes and 355 Todo scopes.
Read all deferred facts with the original capture and identity. Implemented The linked cursor and the historical selector cover the same facts. The saved projection includes checkpointed approvals.
Recover exact scope when retention is lost. Implemented Warm tests reuse the observed Goal list and recover the full remainder after compaction. The original admission receipt remains unique.
Keep the ordinary reader when the control rows fit. Implemented The wider first-page fixture keeps the ordinary cursor. Existing explicit-off and exact selection tests pass.

Validation

Tested revision a0133c2da3a2504b668b89896643b94b04e1c3d2 against main at 8e9a28a866da06cc6fd63e78709910fe35dad564. The input classes were synthetic and public-safe.

Check kind Result Evidence
real_entrypoint passed quota should-run and saved quota context-project with real isolated File and SQLite authority. The regression covers cold reads, retained Goal scope, lost retention and checkpointed approvals.
regression_parity passed The same File case fails the generated-opening assertion on base and passes on head. Exact continuation and recovery preserve both scope values and the single admission receipt.
integration passed 37 Python tests for first-read navigation and exact selection transport, including 24 ready-opening cases.
unit passed 28 control-plane selection and progressive-context TypeScript tests.
static passed TypeScript typecheck, changed Python Ruff, public boundary scan, RFC diff check and 19 standard premerge checks.
integration not_run Full local Python suite and installed Host and model qualification. Hosted checks and independent review are pending.

Matched cold File fixture with a 12,000-byte first-page limit

Observation Base Head
Complete first-page selection No Yes
First reader stdout bytes 12,000 6,848
Full initial-read and continuation stdout bytes 67,439 68,963
Calls to read the full initial selection 6 7

The extra call separates the bounded opening from the required scope read. These are CLI wire measurements. Installed Host retention and model cost remain open.

Hosted check follow-up

The submitted head a0133c2da3a2504b668b89896643b94b04e1c3d2 hit three independent checks. The other red jobs report skipped or failed upstream jobs.

Check Failure on the submitted head Existing repair and local verification
Frontstage Pages Strict MkDocs aborts on the runtime guide link outside the published docs directory. Job The main site and both developer books build in strict mode. The rendered protocol page links to the canonical runtime guide. #9032
Python Tests Kernel lint reports an unused pytest import. The required backend checks and shards stop upstream. Job The CI-scope Ruff command and configured kernel mypy check pass. The composition uses the two identical repaired files carried by #9084. #9082
Release Artifacts Stable runtime preparation rejects the unqualified public default contract. Job The release builder's source-archive probe accepts the composed public quota packet default. #9084

The local composition combines the submitted head with #9084 at 35c9b103540bcd5a26e0be6bb1756a57e67ffa2b and #9032 at 1254f7598cf9a2882c4d8a75e6cb54a433c9c41e. All 10 focused File/SQLite cases pass on that tree, including the combined-scope opening, same-capture recovery after lost retention and explicit-off behavior. The larger composition run stopped on local disk capacity.

The hosted checks still reflect the submitted head. Hosted qualification and independent review remain pending. The source-archive probe covers the default contract used by the release builder. The full release artifact and installed Host and model qualification remain pending.

Frontend / Visual Evidence

UI impact is none.

Type of Change

  • Bug fix
  • Test update

LoopX Area

  • Control plane

Commits carry DCO sign-off. Validation disclosure guidance

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 · gpt-6.2-sol · OpenAI (runtime_reported, max)

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

动机 / Motivation

使用原生配额 reader 的 Agent 需要在写入前同时看到就绪、停止条件和必读项。 12,000 字节预算下,旧 reader 会先交付整组 Todo 写入范围。File、SQLite 的 290 项 Goal 范围加 355 项 Todo 范围在冷、暖共四个同条件用例中都没有生成完整的就绪首读。 File 与 SQLite 都在完整首读拿到就绪、停止条件和必读项。同一摘要绑定的续读交付两个范围和审批条目。四项基线回归均通过,24 项宽范围组合全部通过。 验收交付的是当前决策的读取顺序、范围和出典。 本地实际 CLI、已保存决策和模拟 Codex lifecycle 覆盖同一捕获的恢复。已安装 Host 的模型可见历史、实际读取率与完整成本仍待验证。

改动思路 / Approach

保留现有 TS 选择器和 Python reader。向既有开篇分组传递最多两个待拆分对象,发布时继续由实际序列化预算决定是否拆分。只调整当前 native ready 的阅读导航。读回由原 immutable digest、原串行游标及观察失效回退合成。当普通首读容不下控制信息时,把 Goal 和 Todo 两个作用域明细以及适用的审批记录接到原捕获的必读续段,同时提前当前 execution_preparation。审批摘要和选中 Todo 的文本留在就绪首读。

具体改动 / Changes

精确 head a0133c2da3a2504b668b89896643b94b04e1c3d2,diff base 8e9a28a866da06cc6fd63e78709910fe35dad564,3 个文件 +93/-39。逐项复核 首屏决策与必读、同捕获续读与恢复、普通游标分支。验收约定

关键代码讲解

  • 同条件对照中,四项主干用例都缺少完整的就绪分段。候选新增和已有的宽范围 24 项都能读回原身份与精确范围。源码
  • 较小的 Goal 范围下同样能处理 355 项 Todo 范围。增大有效读取预算到 64,000 字节时复用原普通游标。源码
  • 暖读只复用经过观察的 Goal 范围。失去观察后,原续读恢复完整选区,同时保留原准入回执。源码
  • 显式 off 路径用两种 provider 各走一次完整 CLI。规范化后的决策、停止、必读摘要及权威数组逐项一致。源码

对主干的风险 / Risks

扩展 Python 套件 285 项通过,四项既有失败。另通过 4 项独立窄 Goal/宽预算对照、64 项 TypeScript 测试、TypeScript typecheck,以及 5 项直接和 14 项选中标准预合并检查。 扩展 Python 套件的四个既有 Host 用例在基线和候选以同一命令复现,均由 extract_turn_authority 在解析输入时拒绝不支持的 TurnEnvelope 版本。 某些输入增加一次明确续读,整个交付仍须读完。

语义与 CI 对齐

首读到 EOF 后仍须完成续读。新增内容只把本次请求范围的具体位置写入文档。同一 capture 的 owner、执行授权和停止条件继续由原契约决定。

我的整体评价 / Verdict

同意此精确提交。它让 Agent 在分段时先看到当前的执行约束,同时保住完成读取所需的原始捕获。调用方有足够预算时,仍可沿普通游标读完。

English verdict: APPROVE. The bounded ready opening keeps execution controls visible with large Todo scopes. Exact-base failures, both authority backends, original-capture recovery, a fitting ordinary cursor, and explicit-off parity were checked. Installed-host and model retention were outside the local evidence.

This branch has not been deployed

No deployments
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.

1 participant