Checked October 8, 2026. This record helps contributors choose useful upstream work; it is not a merge queue, endorsement or adoption count. The ecosystem inventory separately records downstream use and derivative projects.
The public GitHub fork endpoint returned 602 distinct repository records during the crawl. Git branch discovery succeeded for 596; six repositories returned not found. The upstream API reported 596 forks at that observation. These are repository observations, not independently active users or deployments.
The readable repositories exposed 43,033 branch references and 3,344 distinct tips. Comparing them with upstream's current branches, all exposed upstream PR heads and their reachable history reduced the unresolved set to 771 tips across 101 repositories. Unresolved does not mean novel: cherry-picks, discarded PR heads and copied branches can still be semantically equivalent. The upstream comparison covered 6,073 refs and 21,734 reachable commits. The initial metadata-only GraphQL attempt had partial batches and was replaced by complete Git branch discovery; it is not used as evidence of full branch coverage.
Discovery was pinned to d605f73720c48c5f9bc4cbb5bd669352e7dcc709. The concrete source comparisons below also checked b6e0e53a81840059026ace3ef099f546901e2dfe. This is a dated checkpoint, not a claim that a moving branch stays unchanged.
PR lists were read for 12 targeted forks: ApexAiOfficial, Schofi, luning2024, xiaods, liangsalt, elegracer, bmshin94, Sande33p, yanfeng98, lightencc, manoelcalixto and douyipu. Branches without a fork-local PR were assessed as branches, not invented PRs. Known derivatives outside the fork network were checked against the existing inventory and prior contribution conversations. Deep review prioritised concrete current defects and reusable seams; every unresolved branch was not individually reviewed or executed.
| Source and exact revision | Value for current upstream | Decision and remaining qualification |
|---|---|---|
ApexAiOfficial / Logan Pratt: MCP Todo query and CLI resolution (fix/mcp-list-todos-and-cli-resolution) |
Current GoalModeMCPControlPlane.list_todos still calls should_run. A read-only dispatch probe confirmed a quota request, rather than todo list. The quota/work-selection packet can expose selected Todo IDs; replaying a recorded real packet through the production MCP method retained selected_todo.todo_id. The missing contract is the complete native Todo-list query, not all ID discovery. The GUI/PATH case is a second useful portability concern. |
Invite a focused upstream PR. Preserve today's bound GoalRef/agent checks and transport owner. Qualify executable selection and invalid overrides; the fork patch only checks path existence. No fork-local or upstream author PR was found at this checkpoint. The full fork was not compatibility-tested. |
Sande33p: fork-safe Frontstage workflow, PR #3, head 3aab8d882258f587d377787731ab68d9cdcb7876 |
The current upstream workflow gates publication steps by event, but lacks repository-level build/deploy isolation. A small workflow fix can retain fork PR validation while avoiding publication credentials on fork push/schedule/manual events. | Invite an adapted upstream PR. Its historical huangruiteng/loopx literal must become the actual current loopx-project/loopx identity; copying it unchanged would disable upstream publication. Inspect real event/repository counterfactuals and the current workflow smoke. PR remains open downstream; its reported validation was not independently reproduced here. |
Schofi: lazy CLI routing (codex/lazy-cli-dispatch-20260803) |
General lazy group registration is relevant to remaining cold paths. However, upstream already merged PR #3717 and now owns selected-command routing in entrypoint.py / cli_runtime.py. |
Defer wholesale import. Re-measure the same current commands and extend the existing routing owner only for an evidenced remaining gap. The fork's historical timing/module counts are author claims, not current base/head measurements. Its static command table would introduce maintenance alongside today's owner. |
luning2024: Windows process support (codex-windows-support; includes a477a179d1f99fc0d31cda9ec770f649a90faea2) |
Cross-platform launch/drain remains important. Current upstream already has Windows installation and an existing process runtime/host transport boundary; the old 16-file patch targets an earlier layout. | Defer broad transplant. Bring any reproducible remaining Windows failure to the current process owner and run on a real Windows host. No Windows execution or whole-fork compatibility was qualified in this research. |
xiaods: stale Pi user-prompt write (codex/pi-host-v2) |
The change protects a newly activated Goal from an old asynchronous pause. | Already represented upstream; no duplicate invitation. Current userPrompt uses expected generation/Goal identity and only cancels the timer after a successful CAS write. The broader historical host branch is not assumed current-compatible. |
Sande33p: scheduler submission diagnostics, PR #2, head 00656f8d8f89d907105962ed1fb3def0a8b6dc0e |
Private, bounded rejected-submission diagnostics and site-owned PBS defaults are valuable to the optional scientific provider. | Retain as provider-level follow-up. Upstreaming the entire scientific campaign package needs an explicit optional-provider boundary and current install/authority qualification. Downstream PR is open; its body retains a Node-backed canary hold. A reported live scheduler submission is not independent upstream acceptance. |
manoelcalixto: English dashboard copy, PR #1, head 03dafa68c3d5a0cce2eecf230bbf218a31b5d95a |
Downstream localisation was merged there. Current upstream has a bilingual localisation owner. | Do not import old string replacement wholesale. Any current untranslated journey should be reproduced against the existing i18n boundary. A downstream merge is not an upstream merge. |
| bmshin94: Korean study notes | A useful historical onboarding/research perspective, including packaging observations. | Reference, not current operating guidance. The pinned notes explicitly identify Apache-2.0 since v0.4.8 and MIT before it; they do not assume the current licence is MIT. Recheck their historical Python layout, installation paths and version compatibility against today's TypeScript control-plane owner before proposing translation or reuse; preserve attribution. |
The michaelx1993 → foolzzz derivative family remains one family, not two independent adopters. Its deterministic bookkeeping/digest work is a research lead whose present compatibility was not established by this branch scan. needware/foreman PR #1 already has maintainer invitations and the upstream TypeScript coordination issue #3225; another identical invitation would add no value. Existing closed-but-unmerged fork PRs are not counted as delivered upstream work.
- MCP fix: invitation on the exact ApexAiOfficial commit. Published and read back October 8. Author response, upstream PR creation and merge remain separate, unobserved outcomes.
- Frontstage isolation: invitation is recorded on downstream PR #3, published and read back October 8. The proposed upstream adaptation must correct the repository identity and retain fork PR validation. An invitation does not qualify or merge the patch.
Use the original author and precise source revision in any proposal. ApexAiOfficial's inspected revision carries Apache-2.0; the inspected Schofi and luning2024 revisions carry historical MIT notices. The current upstream contribution boundary is Apache-2.0 with DCO, including applicable historical notices. Do not assume a default-branch license proves rights to every added asset or contribution.
Before selecting a candidate again, read its current exact head, upstream equivalent implementation, author conversation and licence at the selected revision. A useful contribution should name the current defect or missing outcome, reuse its nearest owner, and include a regression through the real caller. Record source-only inspection, author-reported tests, independent reproduction, actual upstream PR and merge separately. Recheck current Windows/HPC environments only when a concrete proposal requires them; this inventory did not run fork schedulers, install providers or publish websites.
中文摘要：本轮完整读取可访问 fork 的 Git 分支，再按当前 upstream 分支和 PR 历史去重；未匹配的 tip 不等于独立创新。优先邀请 MCP Todo 查询修复和经组织身份修正的 Frontstage fork 隔离；CLI、Windows 与旧宿主整支移植暂缓，已有实现或已有邀请的候选不重复联系。源码检查、作者测试陈述、独立复现、邀请、upstream PR、合并分别记录。