Skip to content

[PR] fix: read Codex model_catalog_json so custom-catalog installs stop falling back to 272K (#5243) - #5276

Closed
ranxianglei wants to merge 2 commits into
masterfrom
2026-10-31_codex-model-catalog-window
Closed

ranxianglei wants to merge 2 commits into
masterfrom
2026-10-31_codex-model-catalog-window

Conversation

@ranxianglei

@ranxianglei ranxianglei commented Oct 30, 2026 •

Copy link
Copy Markdown
Owner

Duplicate coordination: a second implementation of the same fix landed as #5305 (cites #5243). Both resolve the reported case identically (catalog install → 828400); they diverge only in one edge case — see the thread for the comparison and consolidation recommendation. This PR references #5243, the upstream Codex catalog-window issue. (Upstream #5246 is an unrelated omp-lane PR-tracking issue; this work is tracked as #5246 in the ework thread but deliberately does NOT close it.)

Problem

Codex resolves its live model table from exactly one path, <CODEX_HOME>/models_cache.json. On installs driven by a custom catalog — config.toml sets model_catalog_json = "/path/to/model-catalog.json" — that cache file never exists, so bili could not find the model, fell through to the bundled snapshot (no entry) and landed on the flat CODEX_FALLBACK_CONTEXT_WINDOW = 272000. The effective window got clamped to 272K while codex actually perceives ~828K (max_context_window 872000 × 95%), so the nudge denominator was ~1/3 of reality and EMERGENCY nudges fired at ~35% of the real window. Root cause confirmed against v0.1.209 source.

Two aggravators were also present: parseCodexToml never read model_catalog_json, and the user's explicit model_context_window = 1050000 was discarded because the early undefined return bypassed it.

Fix

  1. Resolve the live table from the catalog path. parseCodexToml now reads top-level model_catalog_json; the live-table resolver uses that path when set, else <CODEX_HOME>/models_cache.json (relative paths resolve against CODEX_HOME). Same stat-keyed cache, same parsing/validation.
  2. Config-only fallback capped by the known ceiling. When neither file is readable but config.toml sets model_context_window, size from that value capped by the known max_context_window (the bundled entry's max, or the 272K unknown-model ceiling) rather than the raw override — so we never budget above what codex perceives — and log a warn naming the degraded path.
  3. Name the winning source. The clamp log line now reports which rung won (client-report / catalog-json / models-cache / config-context-window / bundled-snapshot / flat-fallback) instead of the misleading "current client metadata preferred over bundled fallback".

The issue's exact case now computes min(1050000, 872000) × 95% = 828400 — matching the rollout-recorded real window.

Behavior change (old → new)

  • Local codex peers on custom-catalog / missing-cache installs: effective window rises from 272K toward codex's real perception (never below it); the nudge denominator matches reality.
  • Unchanged: remote peers (still trust client-reported), non-codex clients, operator-set compress.modelContextLimit, the min() semantics, output reservation, and every existing call path — the legacy codexAlignedWindow return shape is preserved verbatim via a thin wrapper, so all existing assertions hold byte-for-byte.

Config surface

No new bili config knob. model_catalog_json is read from codex's own ~/.codex/config.toml as an input source (bili consumes it, never writes it); no BILI_* env var and no ~/.config/billion-context/billion-context.json field is added, so CONFIGURATION.md en/zh need no change.

Tests

New targeted coverage in tests/codex-live-window.test.ts: catalog-path resolution (issue repro → 828400), catalog outranking models_cache.json, config-only fallback capping (not-in-bundled → 258400; bundled-max-1M gpt-5.4 → 950000; bundled-max-272K gpt-5.5 → 258400; no override → undefined), and full source-attribution across every rung including non-codex passthrough. 30/30 across the three codex-window files pass locally; typecheck clean on the rebased head. Full-suite verdict deferred to CI.

Merge tier

Touches the window/fallback default-value decision path → §7.4 human merge required (not auto-merge eligible).

Closes #5243

@ranxianglei ranxianglei changed the title fix: read Codex model_catalog_json so custom-catalog installs stop falling back to 272K (#5246) [PR] fix: read Codex model_catalog_json so custom-catalog installs stop falling back to 272K (#5246) Oct 30, 2026
@github-actions

github-actions Bot commented Oct 30, 2026 •

Copy link
Copy Markdown
Contributor

📦 Built Package Artifact

Branch: 2026-10-31_codex-model-catalog-window (8201c68)

Latest build — install this exact version

npm install -g billion-context@0.0.7298-pr.5276.c64514e

This exact version is immutable and reproducible. The rolling pr tag tracks the newest test build across all PRs.

Recent builds (commit → version)

commit version build
8201c68 billion-context@0.0.7298-pr.5276.c64514e build
480d1f0 billion-context@0.0.7242-pr.5276.232f498 build

Download artifact

  1. Download the artifact from the Actions run
  2. Extract the tarball and install:
tar xzf billion-context-pr5276.tgz
npm install -g package

Install instructions are refreshed on each push.

ci-check and others added 2 commits October 31, 2026 09:27
…lling back to 272K (#5246)

Codex resolves its live model table from exactly one path, <CODEX_HOME>/models_cache.json. On installs driven by a custom catalog (config.toml `model_catalog_json = ...`) that file does not exist, so bili fell through to the bundled snapshot (no entry) and the flat 272K floor, clamping the effective window far below what codex actually perceives and triggering emergency nudges at ~35% of the real window.

- Parse `model_catalog_json` from codex's own config.toml and resolve the live table from that path when set, else models_cache.json (relative paths resolve against CODEX_HOME).
- When neither file is readable but config.toml sets `model_context_window`, size from that value capped by the known max_context_window (the bundled entry's max, or the 272K unknown-model ceiling) instead of the flat 272K floor — never budgeting above what codex perceives — and log a warn naming the degraded path.
- Name the winning window source (client-report/catalog-json/models-cache/config-context-window/bundled-snapshot/flat-fallback) in the clamp log line instead of the misleading "current client metadata preferred" text.

Local codex peers on such installs now get the real perceived window (toward larger, never below codex perception); remote peers, non-codex clients, operator modelContextLimit, min() semantics and output reservation are unchanged.

Closes #5246
The sole call site moved to src/server/handle.ts when handle() was
extracted (e7af746); the import at server.ts:17 has been unused since.
Found while reviewing #5246 — trivial cleanup, no behavior change.
@ranxianglei
ranxianglei force-pushed the 2026-10-31_codex-model-catalog-window branch from 480d1f0 to 8201c68 Compare October 31, 2026 01:37
@ranxianglei ranxianglei added bug Something isn't working codex p2 scope/wiring labels Oct 31, 2026
@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)

CI 终态更新(head 8201c681c):真实全量判定 = 我推送触发的 run 41156221208——全部 guard/static-gate 绿,4 个 test 分片红(ubu22-s2/ubu24-s2/win22-s1/win22-s2)。逐条核过 job 日志,两个失败签名均为在档继承家族,与本 PR diff 零交集:① issue4982-draining-twin-ladder e2e 以 StorageSecondWriterError 死掉而非 ladder(#5046 家族,#5236 在档等 A/B/C 裁决);② claude-native hook e2e「healthy proxy on ANOTHER port」多实例启动竞态(#5136 flake 家族)。已在 #4599 留一行备案。另注:其后一个显示 success 的 run 是标签事件触发的 scope=skip(矩阵整体跳过),不能当作全量绿。

收工判定不变:代码审核完成、本地定向全绿(codex-window 三套件 38/38、model-windows+launcher-max-outputs 30/30、launcher 343 过、typecheck 干净);全量红与 #5046 owner 裁决链相关,非本 PR 引入。§7.4 人工合并——建议要么等继承车道收敛后合,要么 owner 明确接受当前继承红直接合(本 PR 触碰面在专属测试里全绿)。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)

收工总结(本线程 = GitHub PR #5276 的跟踪 issue):已完成的工作与结论全部在本线程——floor 3(独立审核八问 + 行为变更披露)、floor 8(CI 终态)。一句话结果:代码审核通过、分支已同步最新 master 并推送(head 8201c681c)、本地定向全绿;全量 CI 的 4 个红分片均为在档继承家族(#5046 裁决链 + #5136 flake),与本 PR diff 零交集,已备案 #4599。当前状态:等 owner——§7.4 人工合并;另见系统提示:PR 落后 master 已由专职守卫接管同步、CI 自动修复车道 24h 窗口已达上限,两者均不需要本车道再动作。

@ranxianglei ranxianglei changed the title [PR] fix: read Codex model_catalog_json so custom-catalog installs stop falling back to 272K (#5246) fix: read Codex model_catalog_json so custom-catalog installs stop falling back to 272K (#5243) Oct 31, 2026
@ranxianglei ranxianglei changed the title fix: read Codex model_catalog_json so custom-catalog installs stop falling back to 272K (#5243) [PR] fix: read Codex model_catalog_json so custom-catalog installs stop falling back to 272K (#5243) Oct 31, 2026
@ranxianglei

Copy link
Copy Markdown
Owner Author

(ai: sglang-gpu0/qwen3.8-27b)

Superseded — 本 PR 的修复已作为 #5305 合入 master。

#5305(同一 issue #5243)今天 06:55 UTC 已合入,merge commit afdf889cf(即当前 master head)。两实现的差异留档:

  1. 配置窗 fallback:[PR] fix(#5243): honor codex's model_catalog_json when aligning windows #5305 用 model_context_window 原值;本 PR 按 bundled max_context_window×percent 封顶(≤ codex 自身感知值)。合入版取了原值——如希望补上封顶语义,可作为 Codex model_catalog_json installs still fall back to the 272K window — #2600 fix only reads models_cache.json #5243 的小后续。
  2. catalog 不可读时:[PR] fix(#5243): honor codex's model_catalog_json when aligning windows #5305 继续回退读 models_cache.json;本 PR 直接落到 config 窗。
  3. [PR] fix(#5243): honor codex's model_catalog_json when aligning windows #5305 附带了专门回归套件 tests/issue5243-codex-catalog-window.test.ts(9 用例含 issue 原数值 828400 钉)。

建议 owner 关闭本 PR(与 #5305 重复),保留分支作审计即可。CI 红分片归因结论(#5046/#5136 继承家族)不受影响。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)

核实结论:本 PR 确已被 #5305 取代,建议 owner 关闭(agent 不代关)。

我没有照抄 floor 11,而是独立用 GitHub REST API 复核了一遍,结论全部成立:

已做:给本镜像 issue 打了 duplicate 标签(label API 本次可用,floor 5 曾报「endpoint not implemented」已不复现);原有标签 bug/codex/p2/pr/scope/wiring 全部保留。

未做(owner-only):按 2026-10-23 政策 agent 不关闭任何 issue/PR。请 owner 关闭 GitHub PR #5276 —— 与已合并的 #5305 重复:同一症状(自定义目录安装落到 272K 兜底)、同一机制(读 codex 模型表对齐窗口)、#5305 先合入,后到的 #5276 让位。分支留作审计即可,不用删。关闭后 floor 8/10 归因的那几个继承红分片(#5046/#5136 家族)随单作废,无需再追。

留档的一个行为差异(需 owner 定夺,非本单范围):#5305 在「两个文件都不可读、且 config 设了 model_context_window」这一档用的是原值;本 PR 当时的实现是封顶于 bundled max(保证永不超过 codex 自身感知)。若用户显式把 model_context_window 设得比 codex 真实感知还高,#5305 会按偏高的原值定窗(有界、只在该降级档触发)。是否要补上封顶语义属于可选小后续,等你拍板;我不擅自立项、也不改码。

判定:代码侧无新增动作,等 owner 关单即收工。

@ranxianglei ranxianglei added the duplicate This issue or pull request already exists label Oct 31, 2026
@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)
🔶 需要决策 — 本 PR(#5276)已被合入的 #5305 取代,它是「待 owner 关闭的重复 PR」而不是「可救活的冲突 PR」。请拍板处置方式(推荐 A:直接关闭)。

查重/核实(本轮全部实时查证,不复用旧评论结论)

为什么本轮不做 rebase/force-push

  1. AGENTS.md「上游 PR 冲突唤醒」明文例外:分支已废弃/PR 已无需合并时,回复说明即可,不要硬救。本 PR 的全部目的已由 [PR] fix(#5243): honor codex's model_catalog_json when aligning windows #5305 交付,只剩「等 owner 关闭」这一个终态动作。
  2. 两套实现存在语义分叉(floor 15 留档:[PR] fix(#5243): honor codex's model_catalog_json when aligning windows #5305 用 model_context_window 原值定窗;本 PR 按 bundled max 封顶)。rebase 解冲突 = 在两套实现之间做取舍,是未决的设计决策,超出本车道「small mechanical fixes only」权限。
  3. 强推会在全量 CI 矩阵上烧一轮舰队资源,而该 PR 的去向是关闭不是合并;§7.4 上它本就属人工合并档。
  4. 分支已落后 master 至少 1232 个提交(本地陈旧 ref 计,实际只会更多),横跨大量热文件重写,冲突面远超「机械解决」。

选项

红 CI 备注:镜像缓存快照(2026-10-31)里 3 个 test 分片红的归因见 floor 8(#5046/#5136 继承家族,已备案 #4599),与本 PR diff 无关;若按 A 关闭则无需再追。

标签:既有 bug/codex/pr/scope/wiring/p2/duplicate;本轮新增 needs-decision(本工作区没有 add-label.sh,直接经镜像 label API 做并集提交,已验证生效)。

结构化八问

  1. 是什么问题:系统把「已被取代、只待关闭」的重复 PR 当成普通冲突 PR 反复唤醒 CI 车道;正确终态是 owner 关闭,不是 rebase 救活。
  2. 什么原因导致的:[PR] fix(#5243): honor codex's model_catalog_json when aligning windows #5305 先合入后本 PR 被判定 duplicate 并等待关闭,期间 master 持续前进使 merge state 变 dirty;冲突触发按 open+dirty 派单,不区分「superseded 待关」状态,于是连续多日重派同一任务。
  3. 影响范围与严重程度:用户面无影响(修复已随 [PR] fix(#5243): honor codex's model_catalog_json when aligning windows #5305 上线);内部为每日空转唤醒 + 若误走 rebase 的全量 CI 资源浪费,属体验/资源瑕疵级,非功能缺陷。
  4. 怎么处理的:零代码动作——实时查证四事实(open + dirty + head 自 10-31 未动 + [PR] fix(#5243): honor codex's model_catalog_json when aligning windows #5305 已合入且回归套件在 master)后,按「已废弃 PR 不硬救」例外停手,打 needs-decision 标签把处置权交回 owner。
  5. 是否有回归问题:不涉及——无 diff、无 push、无任何对 master 或在途车道的改动。
  6. 是否破坏缓存:不涉及。
  7. 是否改了设计方向:不改。仅留档既有实现分叉(封顶 vs 原值)供拍板;只有选 B 才会产生新的后续小工作。
  8. 是否引入新的配置:不涉及。

收工判定:本轮无代码动作(不 rebase、不 push、不开新 issue);needs-decision 标签已打;等 owner 拍板——建议 A(关闭 #5276),关单后本线程的冲突唤醒自然停止。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)
🔶 需要决策(状态复核,无新动作)— 本轮再次实时查证,状态与 floor 20 的决策请求完全一致,处置权仍在 owner。(查重结论见 floor 20:与已合并的 #5305 同症状同机制,duplicate。)

本轮实时查证(不复用记忆)

结论不变:本 PR 是「已被 #5305 取代、只待 owner 关闭」的重复 PR。rebase 解冲突 = 在两套实现(原值 vs 封顶)之间做设计取舍,超出本车道「small mechanical fixes only」权限;强推只会额外烧一轮全量 CI,而该 PR 的去向是关闭不是合并。选项与推荐沿用 floor 20:A(推荐)owner 直接关闭 #5276(duplicate of #5305);B:关单后另立小后续补封顶语义;C:owner 指定以哪套实现为准后由人主导 rebase。default-if-undecided:按 A 等待关闭(agent 依 2026-10-23 政策不代关)。

七问(本轮)

  1. 是什么问题:同一「PR 冲突」任务被重复派发,但该 PR 实际处于 superseded 待关状态,不存在可执行的机械修复。
  2. 什么原因导致的:触发器按 open+dirty 派单,不识别「superseded + needs-decision 在档」这一状态,于是同一任务反复重派。
  3. 影响范围与严重程度:仅内部资源空转(每次触发消耗一轮 agent 运行);用户面无影响(修复早已随 [PR] fix(#5243): honor codex's model_catalog_json when aligning windows #5305 上线)。
  4. 怎么处理的:零代码动作——实时复核全部事实后确认 floor 20 的决策请求仍然有效,本轮只留一行式状态记录,不重复长文。
  5. 是否有回归问题:不涉及——无 diff、无 push、未触碰任何分支或 master。
  6. 是否破坏缓存:不涉及。
  7. 是否改了设计方向 / 引入新配置:均不涉及。

收工判定:等 owner 拍板(建议 A:关闭 #5276)。在 owner 行动前,本车道每次被唤醒得到的都会是同一个复核结论——关闭 PR 是终止该循环的唯一终态动作。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)
🔶 需要决策(第三次复核,无新动作)— 本轮实时复核与 floor 20/22 零变化,处置权仍在 owner。

本轮查证(实时):#5276 仍 open、未合并、head=8201c681c 自 10-31 未动、closed_at=null、mergeable_state=dirty;#5305 维持 merged;标签集不变(duplicate + needs-decision 均在档);镜像缓存刚刷新(fetched_at=2026-11-11T06:06Z),30 项 check 仍为 3 个 test 分片红——归因沿用 floor 8(#5046/#5136 继承家族,已备案 #4599),CI 红项无需本车道动作。线程内无新增 owner 评论。

结论不变:本 PR 是已被 #5305 取代、只待关闭的重复 PR;解冲突 = 在两套实现间做设计取舍,超出本车道权限。选项与推荐沿用 floor 20:A(推荐)owner 直接关闭 #5276;B:关单后另立小后续补封顶语义;C:owner 指定实现后由人主导 rebase。default-if-undecided:按 A 等待关闭(agent 不代关)。

七问:① 问题:同一任务被重复派发,而 PR 实际处于 superseded 待关状态,无机械修复可执行。② 原因:触发器按 open+dirty 派单,不识别 superseded/needs-decision 状态。③ 影响:仅内部资源空转,用户面无影响。④ 处理:零代码动作,仅实时复核并留一行式记录。⑤ 回归:不涉及(无 diff、无 push)。⑥ 缓存:不涉及。⑦ 设计方向/新配置:均不涉及。

收工判定:等 owner 拍板(建议 A:关闭 #5276)。另注:连续三次派单结论相同,duplicate 与 needs-decision 标签均未使唤醒停止——若需在 owner 行动前终止该循环,需平台侧调整触发条件(本 agent 无此权限,如实报告)。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)
🔶 需要决策(第四次复核,零变化)— 实时查证:#5276 仍 open/dirty、head=8201c681c 自 10-31 未动、closed_at=null;#5305 维持 merged;duplicate + needs-decision 标签在档;镜像缓存刚刷新(06:30Z)仍为同样 3 个继承家族红分片(floor 8 归因,备案 #4599);线程无新增 owner 评论。

结论与选项不变(floor 20):A(推荐)owner 关闭 #5276(duplicate of #5305);B:关单后另立封顶语义小后续;C:owner 指定实现后由人主导 rebase。default-if-undecided:按 A 等待关闭。

七问:① superseded 待关 PR 被重复派单,无机械修复可执行;② 触发器不识别 superseded/needs-decision 状态;③ 仅内部空转,用户面无影响;④ 本轮零代码动作,仅实时复核;⑤⑥⑦ 回归/缓存/设计配置均不涉及(无 diff、无 push)。

收工判定:等 owner 拍板。连续四次派单结论相同——终止循环需 owner 关单或平台侧调整触发条件。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)
🔶 需要决策(第五次复核,零变化)— 实时查证:#5276 仍 open/dirty、head=8201c681c 自 10-31 未动、closed_at=null;duplicate + needs-decision 标签在档;镜像缓存(11:06Z 刷新)仍为同样 3 个继承家族红分片(floor 8 归因,备案 #4599);线程无 owner 新评论,仅系统接管/处理标记。

结论与选项不变(floor 20):A(推荐)owner 关闭 #5276(duplicate of #5305);B:关单后另立封顶语义小后续;C:owner 指定实现后由人主导 rebase。default-if-undecided:按 A 等待关闭。

七问:① superseded 待关 PR 被重复派单,无机械修复可执行;② 触发器不识别 superseded/needs-decision 状态;③ 仅内部空转,用户面无影响;④ 本轮零代码动作,仅实时复核;⑤⑥⑦ 回归/缓存/设计配置均不涉及(无 diff、无 push)。

收工判定:等 owner 拍板。已连续五次派单、跨 11-11 至 11-12 三天,结论恒定——终止循环需 owner 关单或平台侧调整触发条件。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)
🔶 需要决策(第六次复核,零变化)— 实时查证:#5276 仍 open/dirty、head=8201c681c 自 10-31 未动、closed_at=null;duplicate + needs-decision 标签在档;镜像缓存仍为同样 3 个继承家族红分片(floor 8 归因,备案 #4599);线程无 owner 新评论。

结论与选项不变(floor 20):A(推荐)owner 关闭 #5276(duplicate of #5305);B:关单后另立封顶语义小后续;C:owner 指定实现后由人主导 rebase。default-if-undecided:按 A 等待关闭。

七问:① superseded 待关 PR 被重复派单,无机械修复可执行;② 触发器不识别 superseded/needs-decision 状态;③ 仅内部空转,用户面无影响;④ 本轮零代码动作,仅实时复核;⑤⑥⑦ 回归/缓存/设计配置均不涉及(无 diff、无 push)。

收工判定:等 owner 拍板。连续六次派单结论恒定——终止循环需 owner 关单或平台侧调整触发条件。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)
🔶 需要决策(第七次复核,零变化)— 实时查证:#5276 仍 open/dirty、head=8201c681c 自 10-31 未动、closed_at=null;duplicate + needs-decision 标签在档;镜像缓存仍为同样 3 个继承家族红分片(floor 8 归因,备案 #4599);线程无 owner 新评论。

结论与选项不变(floor 20):A(推荐)owner 关闭 #5276(duplicate of #5305);B:关单后另立封顶语义小后续;C:owner 指定实现后由人主导 rebase。default-if-undecided:按 A 等待关闭。

七问:① superseded 待关 PR 被重复派单,无机械修复可执行;② 触发器不识别 superseded/needs-decision 状态;③ 仅内部空转,用户面无影响;④ 本轮零代码动作,仅实时复核;⑤⑥⑦ 回归/缓存/设计配置均不涉及(无 diff、无 push)。

收工判定:等 owner 拍板。连续七次派单结论恒定——终止循环需 owner 关单或平台侧调整触发条件。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)
🔶 需要决策(第八次复核,零变化)— 实时查证:#5276 仍 open/dirty、head=8201c681c 自 10-31 未动、closed_at=null;duplicate + needs-decision 标签在档;镜像缓存(12-13 刷新)仍为同样 3 个继承家族红分片(floor 8 归因,备案 #4599);线程无 owner 新评论。

结论与选项不变(floor 20):A(推荐)owner 关闭 #5276(duplicate of #5305);B:关单后另立封顶语义小后续;C:owner 指定实现后由人主导 rebase。default-if-undecided:按 A 等待关闭。

七问:① superseded 待关 PR 被重复派单,无机械修复可执行;② 触发器不识别 superseded/needs-decision 状态;③ 仅内部空转,用户面无影响;④ 本轮零代码动作,仅实时复核;⑤⑥⑦ 回归/缓存/设计配置均不涉及(无 diff、无 push)。

收工判定:等 owner 拍板。连续八次派单、跨 11-11 至 11-13 四天,结论恒定——终止循环需 owner 关单或平台侧调整触发条件。

@ranxianglei

Copy link
Copy Markdown
Owner Author

Closing: the fix landed on master as 0629480 "fix: honor codex's model_catalog_json when aligning windows (#5243)" (model_catalog_json read in src/loop/adapter-codex.ts) plus 73e6d2b (export hygiene).

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

bug Something isn't working codex duplicate This issue or pull request already exists needs-decision p2 scope/wiring

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Codex model_catalog_json installs still fall back to the 272K window — #2600 fix only reads models_cache.json

1 participant