Handoff: the state of the work

How to use this file

This file records where the work on ngx_mruby stands, so that any session, clone or contributor can continue it: the open pull requests, the decisions that are not written down elsewhere yet, the work in progress and the queue.

• It records state, not rules. The rules are in AGENTS.md. The plan for v3 and the owner’s decisions about it are in docs/proposals/v3-plan.md. Where this file disagrees with them, they win, with the one exception that AGENTS.md “Session handoff” sets: a dated row under “Decisions” holds over their older text until a pull request writes it there, but only when the row links the owner’s comment or quotes the owner’s words. A row with neither is a proposal of the sessions and changes no rule.
• A decision of the owner that is not written there yet is recorded under “Decisions”, in short, with its date, the link or the quotation, and the place where it will be written. Where the sessions apply a decision with a procedure of their own, the row names that procedure separately.
• A pull request on next updates this file in the same pull request when it changes what the file records, its own row and the queue item that it finishes included. A pull request on master cannot, because the file is only on next: the session that merged it into master opens a small pull request on next that updates the file, or updates the file in the merge of master into next if that comes first. A release is recorded in the same way by the first session that finds its tag missing here. Link pull requests and issues instead of copying their text.
• Remove an item when its pull request is merged or closed, when its issue is closed, or when it is written in the place that its row names. The pull requests and issues keep the history (see the gh commands below).
• The file lives on next. From a checkout of another branch, read it with git fetch origin next and git show origin/next:docs/HANDOFF.md.
• The statements about branches, pull requests and issues were checked with gh and git at the time in the next heading. Check them again before you act on them: other sessions merge pull requests all the time.

Whole-page material integration, 2026-10-30

• #668 merged as 252018d86; production run 41023287679 succeeded. Its 626 shipped files matched production; live code-layout verification covered 340 conditions and interaction verification 32. The earlier pending-publication entries below are historical checkpoints.
• The owner found that metallic components looked detached from the remaining page: background, composition, light, material, color and motion need one system. #669 (codex/home-material-system) addresses that feedback on the v3 overview, preserving tested routing, version states and canonical code copies. The site plan records the scope. Visual approval is separate from passing automated checks.
• This iteration changes homepage styling and its material checks only. No runtime change, MCP interoperability claim, stable README migration, new OG artwork or software release is included. Personal photos remain unpublished.
• Local preview checks passed: 40 homepage style loads, 500 code conditions, 32 interaction conditions, 69 complete exports and English/Japanese accessibility 100. An independent visual review found no must-fix and one arrow-placement improvement; the MCP entry arrow now points at Ruby’s column. A technical review also found transient text fading during route motion. Test-only commit 52cb29ce7 reproduces it (opacity 0.6 at animation start); the correction keeps text opaque, and the interaction check now covers both themes and four points of the animation. The narrow MCP diagram retains the Ruby-to-backend branch layout. Final-head CI and re-review are recorded on the draft PR. A further visual review removed diagonals that crossed content and the nested outer/Ruby cards, retaining a single nginx enclosure and a section-aligned light field. The new visual diff needs final-head verification. Publication remains separate; #668’s individual merge approval is not reused here.

Final site validation follow-up, 2026-10-30

• #668 head 77f3161ad passed test/examples CI, but site CI found an overly wide sample caption and a Japanese identifier-search failure. Publication remains held until the follow-up head passes all checks.
• The caption now uses a character-based width. Search metadata is generated from canonical API names; visible prose/code and exported Markdown stay unchanged. At regression-test commit 3ec909597, the exact-reference check fails only for Japanese mruby_content_handler (11/12 pass). The new test also requires the target reference, not merely any result. Final-head execution and independent review are recorded on #668.
• #667 production run 41011744554 and deployment 7432116276 succeeded; 80 changed/new shipped files matched HTTPS, and the live policy/redirect checks and all five Lighthouse categories (100) passed.
• Work continues in the existing repository checkout. Previous worktrees and branches remain preserved; local folder inventories are not public artifacts.
• The owner’s 2026-10-30 correction requires code to be readable without horizontal scrolling and a modern code presentation. #668 now applies soft wrapping to both versions and all documentation, names sample files, and highlights the model-routing assignment. Full configuration remains expandable; syntax and original clipboard bytes are preserved. Regression commit d88c1e573 fails on the old homepage at 320px. check-code-layout.mjs checks every generated page at 320px and required English/Japanese reference and entry pages at 320/390/768/1440, with 200% code text, unbroken URLs, keyboard activation and canonical sample copies. Local code-layout checks pass in 500 page conditions in Chromium/WebKit, including canonical clipboard and keyboard checks; homepage computed-style checks pass in 40 light/dark page loads. Publication remains pending final-head CI, independent review and live QA.
• Site CI at d005d45f3 passed the complete computed-style, Japanese, focus, search and accessibility gates. Its interaction check counted Starlight’s idle-loaded Pagefind UI module as illustration traffic. The check now waits for the search input to be attached before measuring route controls, retaining the zero-request assertion. A delayed-idle diagnostic reproduces the old failure; normal and delayed initialization each pass all 32 interaction conditions with the wait. Final CI is required.

v3 homepage redesign checkpoint, 2026-10-30 06:04 UTC

• The owner prioritized site expression over further internal C work: v3 as the main entry, concrete users and purposes, large diagrams, metallic surfaces and deliberate motion. The site plan records the revised visual scope and bounded exceptions. Stable 2.7 keeps its own entry and sharing card. Personal photos remain references only.
• #668 owns this work. The new homepage shows tested model routing, an explicitly unverified MCP document-service design, and trial/edit/check entry points. English and Japanese canonical guides feed HTML and machine-readable Markdown.
• #663 production run 41005252954 completed: 117 changed/new shipped files matched over HTTPS; deployment 7431188963, redirects and site policy checks passed, with all five live Lighthouse categories at 100.
• #665 merged as 2677db219 after current-head CI and independent review; production run 41008268601 passed, with 165 changed/new shipped files matching HTTPS. #666 merged as d2827d72c and production run 41009919981 passed. #667 merged as 79d7e3933 after final-head CI and independent review; production verification is pending. This branch includes both.
• A fresh linux/arm64 trial of examples/start stopped at Ubuntu image-metadata retrieval before compilation. No startup/edit/reload success is claimed for that attempt. The updated example adds executable configuration validation and reload checks. Examples CI 41009377947 executed the nine commands successfully on the Linux runner, including build, response changes, validation, reload and cleanup. Final-head CI remains required before publication.
• Internal fixes remain separate in #664 and the existing local checkpoints. The owner accepted the v3 setter choices D/E on 2026-10-30 with 「すごい！両方進めて！」 in response to the explicit proposal: invalid Strings raise ArgumentError and the HTTP setter refuses hostnames without resolving them; non-String TypeError remains. This decision does not apply to 2.7, settle other open proposal questions, or authorize a software release. Internal implementation is deferred while the site work is completed.
• Local validation includes 24 Chromium/WebKit route-control and reduced-motion loads, Chromium clipboard checks, 120 Noto style loads, 1044 focus stops, six Lighthouse pages (all categories 100), and native/Noto Japanese trial and MCP typography checks. Final-head CI and live artifact/OG checks are recorded on #668; these local results do not claim production verification. MCP execution, actual social-network card fetches and the failed arm64 container trial remain unverified.

Standalone routing recipe, 2026-10-30

• This independent documentation unit includes next 490ea459c after the installation integration #663; runtime inputs are unchanged from the originally tested 6d65a12ec. The English and Japanese routing guide now contain a complete loopback configuration, Ruby table, executable checks and failure diagnostics. No module or existing integration fixture changed.
• The exact shell block was extracted from the guides and executed on Ubuntu 22.04 aarch64 as an unprivileged user, using nginx 1.31.8 built from 6d65a12ec and its default gem lock. Five requests passed: known tenants a/b, absent header, unknown key and an address-shaped key. Each checks HTTP status, content type and every body byte, including the path/query and trailing newline. The instance stops using its own prefix.
• An external negative-check copy with a deliberately wrong expected backend returned 1 and removed the instance’s PID file. That altered expectation is not in the guide. The source/code review found no open must-fix or should-fix item. Browser and export validation are recorded with the completed local commit; no remote CI or publication is claimed.
• The tested executable is a development build, not a v3 release. This recipe was not run on v2.7.2, with dynamic module loading or on non-Linux systems. No performance measurement or full module-suite rerun was performed for this documentation-only unit. The owner authorized the normal PR/CI/publication process on 2026-10-30; final-head CI and deployment verification are pending.

Surface continuation, 2026-10-30

• This homepage-only change adjusts the copper-red CTA and mobile spacing. The site plan records color roles, contrast and measured viewport differences. Existing style expectations are preserved; the new contrast assertion is additive.
• Independent source review has no must-fix or should-fix findings. Local checks cover 80 native/Noto style loads, 1032 focus stops and no contrast failures. Those preview builds were dirty; final-head CI and production verification remain required.
• #665 contains the separate README/OG change. #663 merged as 490ea459c; this branch includes that integration. Publication resumed with the owner’s authorization on 2026-10-30; software releases and personal reference photo publication remain excluded.

Sharing image continuation, 2026-10-30

• The initial site/rebrand in #661 is published at next 6d65a12ec; its completion record links the deployed artifact and live checks. #662 aligned maintenance process docs at master 967b61cc6 (mirrored to v2.x).
• #663 integrates the maintenance docs and verifies the development installation walkthrough. It merged as 490ea459c after final-head CI and independent review; production run 41005252954 is in progress.
• This sharing-image change uses the selected metallic mark in one product sharing image for the README and both site versions. The public source, render command, metadata checks and limits are in the site plan. Existing header/favicon derivatives remain; old image URLs are preserved. Local static checks and full-size/300/390-pixel visual checks pass. Final independent source/visual review has no must-fix or should-fix findings. Current-head PR CI and deployment remain pending; this is not a release change.
• The later site work remains distinct: restrained surface/CTA refinements, deeper verified HTTP/stream examples and contextual performance records. The v3 setter work follows those site units under its existing plan.

Installation and process continuation, 2026-10-29

• The first rebrand and versioned-documentation iteration is published. #661 merged as 6d65a12ec. Production run 40959681713 and deployment 7425112363 succeeded. All 614 shipped files match the live site; four versioned English/Japanese searches, mobile/desktop views, code interaction and HTTPS checks pass. Final evidence is in the completion record.
• The common continuation procedure from #659 reached master as 967b61cc6 in #662, after all CI and independent review passed. The v2.x mirror succeeded. HANDOFF remains on next, and master links that source. Branch-specific test process/port handling is preserved; a common workflow does not imply identical harness capabilities.
• #663, branch codex/integrate-stable-install, merges the installation correction of #660 from master 967b61cc6 into next 6d65a12ec. The conflict resolution selects next development source and preserves the v3 core-gem, lock and removed auto-ssl contracts. English/Japanese commands are identical. A fresh clone of 6d65a12ec built and installed nginx 1.31.8 on Ubuntu 22.04 aarch64; the exact documented HTTP trial passed both response assertions, configuration validation, reload and quit. These are documentation changes, with no measured binary input changed. Current head, final CI and independent review belong in the integration PR.
• Next site work: shared README/social branding and controlled metallic surface/CTA refinements, then deeper executable HTTP/stream examples and reproducible performance records. Keep human and agent output on the same sources; retain implementation/planned/under-design distinctions. The upstream setter D/E work and stable empty-filter follow-up remain separate runtime items. No new runtime contract or software release was decided by this documentation integration.

Site continuation checkpoint, 2026-10-29 20:25 UTC

• next is 2588cfd26, the merge of shared-process PR #659. Its final head 71d136799 passed test run 40946099772 and site run 40946099800 against the current base. The independent review has no unresolved findings.
• Integration #658 is merged. Production run 40945526052 deployed 258c852fa; the live home matches its artifact, HTTPS and the verification file pass, the environment permits only next, and Lighthouse categories score 100.
• master is da0d1172d, the merge of installation-guide #660. All final-head CI passed against master 7d81d417a. Its released v2.7.2 / explicit nginx 1.31.8 static build and HTTP lifecycle passed. Its documentation still needs the normal master-to-next integration; do not reimplement stable changes on next. The v2 site build reads the merged master source.
• This PR, branch codex/site-technical-design, introduces the technical home pages, shared navy/graphite palette, metallic C logo family and versioned Markdown/JSON exports. Its design and source rules are in the site proposal. The owner authorized iterative publication on 2026-10-29; this does not authorize a software release or publication of the reference photos. HTTP gold / stream purple is the current design assignment, with labels; LLM proxy is explicitly development with implemented/planned features.
• Generator fixtures, 67 complete Markdown bodies and source metadata, contrast, 26 v2 HTML pages and four version-specific searches pass. Independent retrieval reconstructed the upstream contract and the stable installation steps without confusing versions. Independent source and representative desktop/mobile/favicon image review has no open findings. The first logo-family browser run found a body-font mismatch and a short Japanese line; implementation fixes are committed. The corrected full browser run and final PR CI are in progress; consult the PR for final results, current head and the deployment record.
• Next site steps: complete that validation, merge under the normal gates, verify the live artifact and search, then iterate on deeper examples, README brand adoption and a dedicated social-sharing image. Runtime upstream setter D/E and the stable empty-filter follow-up remain separate queue items. No runtime implementation changed in this site PR.

Continuation checkpoint, 2026-10-29 20:05 UTC

This supplements the integration snapshot below. Verify branch and PR state again before acting; the PRs hold final-head CI and review evidence.

• master and its v2.x mirror are 7d81d417a. Site cutover PRs #655, #657, #656 and #654 are merged. Production run 40928646131 was verified against its artifact, including all 26 /v2/ pages, version-specific English/Japanese search and mobile rendering.
• #658 merged into next as 258c852fa at 20:02 UTC after explicit owner authorization. Its final head ee0025f5d passed CI on merge ref 7c0c9ee0a, including matrix, sanitizer, valgrind, soak, nginx tests, examples and site build. The performance run 40934491055 has 15 ok rows, largest increase +0.21%; the final numeric review is recorded on the PR. Post-merge site publication still needs its own verification.
• #659, codex/shared-handoff-process, records the common Claude/Codex/human continuation procedure. Head cb3d35164 passed CI (the first OpenSSL download failure was resolved by a rerun). This update includes the new next base, so CI must pass again on the updated head. Its only conflict was this dated state record; the integration snapshot is retained below.
• codex/site-technical-design holds the rebrand and machine-readable documentation preview, still uncommitted at this checkpoint. Preserve its task worktree. The export has 51 development and 16 stable source pages, complete-body/source/metadata checks and independent retrieval review. The pre-logo revision passed 440 browser loads and 1,008 focus stops; new global colors and the metallic C logo need final browser QA. No reference photograph is a public asset. The owner authorized iterative site publication on 2026-10-29: “ossのサイトはデプロイしまくっていいのでどんどんやろう”. This does not authorize a v3 software release or publication of private materials. The latest design and source-state details are in the site proposal and the forthcoming design PR.
• #660, head 4723905ef, fixes the stable installation guide. A released v2.7.2 checkout with explicit nginx 1.31.8 passed the documented static build, install and loopback HTTP lifecycle. Custom/dynamic alternatives are explicitly not additional executed results. CI and final review remain on that PR; this is separate from site styling and v3 runtime work.
• Upstream setter D/E and the stable empty-filter follow-up remain queued. The integration prerequisite is now merged. Existing v3 planning decisions hold; the site’s future-facing description must distinguish implemented, planned and still-undecided behavior.

State as of 2026-10-30 (03:49 JST)

Before this integration, next is e469b259c, the merge of #657, and master is 7d81d417a, the merge of #654; v2.x mirrors that commit. This integration brings that master into next. It carries #646 (HTTP content sleep and redirect exception handling), #648 (setter documentation), #650 (release note guidance), #651 (stream sleep outcomes), #652 (setter type, zone and whole-name checks), #642, #655, #656 and #654. The 2.7.3 behavior entries come with their English and Japanese files; these fixes are not reimplemented.

The conflicts were documentation only: retain the removal of auto-ssl and of the old API-reference sorting TODO, keep the next valgrind and test lifecycle rules, add the new test ports and the still-blocking HTTP setter DNS lookup. The mruby 4 initializer arity stays MRB_ARGS_REQ(1). The new HTTP ports 18133–18135 and stream ports 12374–12378 do not collide with next and work with its offset staging. The readiness check now probes 57 HTTP ports. The Linux static suite on nginx 1.31.8 passed at 6bc885fd4, both with the ordinary ports and with TEST_SH_PORT_OFFSET=240: 40 suite reports and 467 assertions each, no failure or crash. The later commits change documentation only. The integration PR records its final-head review, matrix, sanitizer, valgrind, soak, nginx-tests, site and performance results; the merge conditions of AGENTS.md apply to it.

The site cutover is live. #653 added deployment from next; #655 fixed master’s documentation links; #657 added /v2/; #656 added the master validation and dispatch workflows. The first deployment with /v2/ is run 40923881194, from next at e469b259c (deployment 7420103202). Its full site checks passed. The live home page matches its artifact, HTTP redirects to HTTPS, the verification file is unchanged, the deployment environment permits only next, and live Lighthouse categories score 100. Every /v2/ HTML page answers 200 with the artifact’s bytes. Four live Chromium searches (2.x and 3.x, English and Japanese) use only their own version’s index and results; the English and Japanese /v2/ pages have their banner and fit at 390 px. #656’s dispatch started run 40924294520, which passed and was checked against the live artifact as well. #654 merged at 7d81d417a after all final-head checks and independent review. Its dispatch started run 40928646131, which passed and published deployment 7420858976 from next, with the 2.x pages from master at 7d81d417a. All 26 live /v2/ HTML files again matched that artifact; all four live searches, the two mobile checks and check-live.mjs passed. The master push checks and the v2.x mirror also passed. The 3.x development banner and the owner’s go for Q9 remain outstanding; no Discussions/wiki setting was changed by this continuation.

Earlier completed work is listed below for context. For current branch heads and pending work use this snapshot and the queue, then fetch again.

2.7.2 was released on 2026-10-29: the tag v2.7.2 is on d5ae75547, the merge of #624 into master, and the release was published at 13:26 JST. Its release notes are docs/releases/v2.7.2.md and docs/releases/v2.7.2.ja.md, on next with the same content since #628. 2.7.1 was released on 2026-10-19: the tag v2.7.1 is on ce5eb11, the merge of the 2.7.1 changes into master. Its release notes are docs/releases/v2.7.1.md and docs/releases/v2.7.1.ja.md. The release notes of the next 2.x release are docs/releases/v2.7.3.md and docs/releases/v2.7.3.ja.md, created on master by #636 and on next with the same content by the merge of master at c6d3b3b9b.

Merged on 2026-10-29 (JST) after 14:40, into next: #633 (this file and the comparison of the 3.0.0 release notes with 2.7.2; merged at 14:51, 2c96302b7), #635 (test.sh stops only the nginx that it started, by its pid; merged at 16:09, c4e6ac6cc), #634 (valgrind fails the cell; one cell runs under valgrind on pull requests and every cell except the sanitizer cell nightly; 16:51, 8bff5fbd7), #626 (the agent proxy endpoints diagram without a line in its narrow arrangement; 18:11, a9792b82f), site step 5, #629 to #632 (18:39 to 19:31, 12202b8c4 to 5e9d50663; the site row of “Queue” lists them) and #637 (the v3 plan, the archive of two proposals and this file; 19:50, c0df45534) and #641 (TEST_SH_PORT_OFFSET; 20:31, be559a017) and #640 (the auto-ssl mrbgem and NGX_MRUBY_AUTO_SSL removed from next; 21:33, 67ad9b919). Into master: #636 (the auto-ssl gem tested with Redis, Pebble and dehydrated; 16:40, 22f13686f), #639 (CONTRIBUTING.md and other documentation fixes; 17:59, eb9055584), #638 (nginx 1.31.8 and 1.30.6; 18:31, 9acb75949) and #645 (the stream Async.sleep timer cancelled when the session’s pool is freed, and the fiber registered with the GC once per fiber; 20:06, c6d3b3b9b).

Merged on 2026-10-29 (JST) after 13:20, into master: #624 (the 2.7.2 release: #594 and the other changes of the release; merged at 13:25, d5ae75547; GitHub marks #594 merged with it) and #627 (the two sentences of README.md “Branch strategy” in the present; 13:52, 93432cf69). Into next: #625 (this file; 13:40, 2969426f6) and #628 (master at 93432cf69 into next; 14:26, c9c49023e; its description lists the conflicts, how they were resolved, and the test ports of master that it moved on next).

Merged on 2026-10-29 (JST) after 01:20, into next: #615 (master at 6b27f08da into next; merged at 01:46, 216df0ceb), #606 (mruby-redis in the default build of next again, pinned in build_config.rb.lock at 8d9391049; 02:11, 5ca7c955b), #614 (the Pagefind result titles at a size of the scale, and the search check with the results loading and loaded; 03:38, 571212d58), #600 (the OpenAI Responses API in the mock LLM upstream; 05:30, 0d1b3d756), #617 (site step 3; 05:58, dbe4acf5b) and the six parts of site step 4, #618 to #623 (12:02 to 13:13, d5548db41 to 69b245bdb; the site row of “Queue” lists them). Into master: #616 (the check of the 2.7.2 release notes against every pull request since 2.7.1, and the performance comparison with 2.7.1; 02:35, f70b24031).

Merged on 2026-10-29 (JST), into next: #607 (site step 2: the design tokens, the checks of step 2 and the specimens of 4.12; merged at 00:12, 0e7fba441) and #605 (Nginx::Debug.stats and Nginx::Debug.gc tested in the soak; merged at 00:46, 5999a0cc7). Into master: #612 (the entry of the 2.7.2 release notes for Redis.connect_set_udptr and Redis.new without arguments, matsumotory/mruby-redis#109; merged at 00:10, cfe77a506), #603 (tests of the directives and Ruby methods that had none; merged at 00:46, baea78cd7) and #613 (actions/upload-artifact 7.0.2; merged at 01:08, 6b27f08da). In matsumotory/mruby-redis: #109 (the connection of Redis.connect_set_udptr kept in the mrb_state, not in mrb->ud; merged at 00:12, 8d9391049 on the gem’s master). Merged on 2026-10-28, into next: #610 (Nginx::Var accessors defined at their first use, and the variables of Connection, Server and Request read in C; merged at 06:02, a0f4b97a2). Merged on 2026-10-20, into next: #608 (this file, the state as of 00:10; merged at 00:16, 39e870c9f); into master: #604 (mruby-redis in the default build again for 2.7.2; merged at 00:29, e77f9a705).

Merged on 2026-10-19, into next: #602 (the nginx-tests workflow: nginx’s own test suite on nginx with and without ngx_mruby; merged at 23:03 JST, 620e65718), #597 (the bundled ngx_devel_kit v0.3.4 with ./update-devkit-subtree v0.3.4; merged at 22:08 JST, 2de1626), #595 (master at 36f4c82 into next, with #589; its description lists the conflicts in src/http/ngx_http_mruby_var.c and AGENTS.md and how they were resolved), #574 (200 with Content-Length: 0 when a content handler sends no body, #200, and an output body filter that ends with an empty body sends the response; its entries are in docs/releases/v3.0.0.md and .ja.md and the migration guide, 7b5a0c4), #591 (site step 1: site/ with Starlight, the generator of docs/, the checks and .github/workflows/site.yml, a466688), #590 (master into next in two merge commits: master at 4091cbd, with the 2.7.1 changes, #581, #586 and #587, and then master at 03aad70, with #588; its description lists the conflicts and how they were resolved, and it makes the texts of #584 and #586 on a return inside a block one text for 3.0.0 and 2.x) and #585 (the owner’s decisions of 2026-10-15 on the site proposal: section 7 of docs/proposals/site.md records them, section 4 derives the design from the source of three of the owner’s sites, 3.5 plans the diagrams and samples, section 6 says what each step produces and what the owner reviews, and contrast.py checks a second palette, 75bfa6f). Into master: the 2.7.1 changes (ce5eb11), #581 (the test harness runs nginx -s quit with detect_leaks=0, edc85f0), #586 (the note on a return inside a block of handler code in 2.x, 5e7d57e), #587 (a change of the 2.7.1 release notes, 4091cbd), #588 (tests: the request body that a log handler or a filter sees after nginx ended a body read, and the status 500 after an exception in an output header filter; and two sentences of docs/directives/README.md on such a handler, 03aad70), #589 (Nginx::Var rejects names and setter calls that name no variable, and leaves the name passed to set unchanged; it creates docs/releases/v2.7.2.md and docs/releases/v2.7.2.ja.md and the test fragment test/conf/conf.d/67-var-input.conf; merged at 16:35 JST, 36f4c82), #593 (test.sh waits until the test nginx listens instead of sleeping NGINX_HEATTIME, with test/t/cases/_wait_for_listen.rb and an entry in docs/releases/v2.7.2.md; merged at 17:16 JST, 188d324), #592 (README.md and docs/README.md say next to the abstract of the paper what a worker keeps between requests; merged at 19:10 JST, a6c32a4), #596 (the entry of docs/releases/v2.7.2.md and .ja.md for the builds that enable mruby-redis, which get hiredis v1.5.0 since matsumotory/mruby-redis#108; a500846), #599 (the supply chain checks in CI: zizmor, OpenSSF Scorecard and the heads of the GitHub gems; merged at 20:43 JST, ee0a57f), #598 (the trust boundary of the docs: trusted configuration code, untrusted client input; merged at 21:35 JST, 5eec909), #601 (the callgrind performance lane and wider sanitizer log checks on master; merged at 22:33 JST, 68032bb6a), #609 (the Nginx::SSL examples of the docs look the SNI name up in a table; merged at 23:20 JST, 2135d68a4) and #611 (no checkout token kept in the test.yml jobs, and zizmor audits test.yml again; merged at 23:58 JST, 0d09b5054). Issue #200 was closed at 19:42 JST, after #574. #563 was closed on 2026-10-19: its change is part of 2.7.1 (its last comment). #502 was closed on 2026-10-19 after #586; its test and its entry in the 3.0.0 release notes came with #584. Merged on 2026-10-15, into next: #584 (the regression test of #502, test/t/cases/return_in_block.rb on 18117, its entry in the 3.0.0 release notes and the migration guide, and the owner’s decision on the frame of handler code as item 14 of section 7 of the plan), #583 (the proposal docs/proposals/handler-code-frame.md for #502, and one sentence in the private entry of the 3.0.0 release notes and in the migration guide: a top-level def in handler code stays public, 982a20e) and #582 (the release notes for 3.0.0 of the pull requests merged before docs/releases/v3.0.0.md existed, Set#flatten_merge, the migration guide and docs/releases/v3.0.0.ja.md, acdc82f). Merged on 2026-10-14, into next: #572 (the fixed symbol names interned once, bc2488a), #571 (the site proposal docs/proposals/site.md, 570cdab), #569 (this file, e0dfdc6), #561 (mruby 4.0.0, 0b76bfa), #577 (the proposal docs/proposals/nginx-var-method-lookup.md, 53e8d39), #579 (master at b017496 into next, 7696e1c) and #580 (master at 0a1e8ac into next, dcd3103); into master: #576 (.gitignore and CLAUDE.md, 9be7f02), #575 (the “Why it changed” lines and the Japanese file of the release notes, b017496) and #578 (the files job of CI, test/check_added_files.sh, 0a1e8ac). Merged on 2026-10-04: into next #560, #565 and #573 (master into next); into master #562, #564, #566, #568 and #570. The full lists: gh pr list --state merged --base next --search 'merged:>=2026-10-03' --limit 200, and the same with --base master. Each pull request says what it changed.

Open pull requests checked at 03:49 JST: none. This integration is on codex/merge-master-site-continuation; its PR targets next and is kept draft until the final checks and independent review pass. #654, #656 and #657 are merged. Check again before starting another row.

Decisions

Decisions of the owner that are not yet written in the place their row names: (a) and (b) of 2026-10-04 in the next paragraph. The owner’s choice of 2026-10-29 at 05:34 JST, in a session’s question on who merges the pull request of the 2.7.2 release, tags v2.7.2 and publishes the release (“セッションが進める (推奨)”, “the session goes ahead”), covered 2.7.2 only, and it is carried out: the session that prepared the release merged #624 at 13:25 JST, tagged v2.7.2 on d5ae75547 and published the release at 13:26 JST. AGENTS.md keeps the rule that agents never create tags and never publish releases; nothing is written there unless the owner keeps the choice for later releases. The choices among the options of 4.12 of the site proposal are written in docs/proposals/site.md (#607). The decision of 2026-10-19 on the perf lane of master (port the tests and the CI only, without Nginx::Debug) is written in #601 (its description and docs/test/README.md, which next has since the merge of master at 6b27f08da). The decision on the method lookup of Nginx::Var accessors (the owner’s answer of 2026-10-15 to the question of section 4 of the proposal, in a session’s question: the change to introspection is accepted as a behavior change of 3.0.0; and the confirmation of 2026-10-19, in a session’s question, with the option “その内容で合っている (推奨)” (“that content is right”): (a) with the rule (a2), (d), and accessors in introspection) is written in the Status line of the proposal and in the entries of docs/releases/v3.0.0.md and .ja.md (#610).

On 2026-10-04 the owner answered the recommendations of the analysis of the control flow of Nginx::Async with 「全部推奨通りでいこう、安全網から始めて」 (“go with all the recommendations, starting from the safety net”). Two of its decisions, (a) and (b), are not written in the plan or in the release notes yet, and a third, (c), is amended below: (a) Nginx::Async.sleep in a content handler is supported, so it does not raise and nginx waits for the timer; (b) a rewrite handler that ends with no output and no Nginx.return keeps the 2.x behavior on master, and on next the request goes on to the next phase as with NGX_DECLINED; (c) the rewrite of how resumed handlers end goes into master after 2.7.1. On 2026-10-29 the owner amended (c), in a session’s question, with the option “master は最小の直し、書き直しは next (推奨)”: the rewrite lands on next, in the core stabilization of 3.0.0, and master gets only the minimal fix for the content handler sleep. The amendment of (c) is written in item 15 of section 7 and in the Pillar B row “One fiber lifecycle for HTTP and stream”; (a) and (b) are written in the plan when their work lands, under that row and under the row “Redesign response finalization per phase”.

The other owner decisions of 2026-10-29, all in session’s questions, are written in the plan: items 1 and 6 and 15 of section 7 (mruby 4.1.0-rc2 now, auto-ssl out of next, the agent proxy rows in v3.0.0-alpha.1, the parallel start of parts (a) and (c) of step 7, the subrequest 500, the fix of the empty output body filter in 2.7.3). Two are not written in the plan: the rule on the gems that the owner maintains is in AGENTS.md, “Other repositories”, with the owner’s words (#637 on next, #639 on master); and the body of the GitHub release v2.7.2 gets the section “Behavior changes: read before upgrading” in English and then in Japanese, option “足す (推奨)” (the session that prepared the release edits it with gh release edit).

The two decisions of 2026-10-04 that were recorded here before are written where their rows said: the mruby steps in item 1 of section 7 of the plan (#561), and the “Why it changed” lines and the Japanese release notes in docs/releases/README.md (#575).

Work in progress

mruby update on next

• #561 (merged, 0b76bfa) bundles mruby 4.0.0. Its lock pins matsumotory/mruby-uname to 50031c8 on the gem’s master (the merge of matsumotory/mruby-uname#1), so no build depends on the gem’s branch mruby-4. The owner accepted the perf FAIL rows that remained after #572: the record on #561. #577 (merged) is the proposal for the var and auth rows; #610 (merged) implements it.
• next bundles mruby 4.0.0 and moves to 4.1.0-rc2 (the newest mruby tag on 2026-10-29, gh api repos/mruby/mruby/tags), then to 4.1.0 when it is tagged (the owner’s decision of 2026-10-29). The choices for that step are in item 1 of section 7 of the plan, and the row “mruby 4.1.0” of “Queue” lists what the step changes.

Issue triage of 2026-10-04

The closed issues (21 on 2026-10-29, among them #225 closed by #566, #320 by #565 and #519 by #562, and #206, #321, #502 and #200, closed on 2026-10-19; #200 after #574): gh issue list --state closed --search 'closed:>=2026-10-04' --limit 200. The reason is in the last comment of each issue, or in the pull request that closed it.

Taken up and still open: none.

Queue

Follow-up	Target branch	Waits for
Merge master into next after each merge into master. Model: #573.	next	This integration brings master at 7d81d417a, including #654, as listed in the snapshot. Its PR records the final validation. The integration adds no behavior changes beyond the imported 2.7.3 fixes.
The output body filter that ends with an empty body sends a zero size buffer marked as memory, and the client gets no byte (Nginx::Filter#output "", pinned by test/t/cases/filter_connection.rb on master). Decided on 2026-10-29 (option “入れる (推奨)”): 2.7.3 gets the same change to ngx_mrb_finalize_body_filter as #574 on next, so that the response is sent with its status and Content-Length: 0, and the owner approved changing the expectation “an empty String closes the connection without a response” of that file. To do: a pull request to master with the test commit, then the fix commit, and the entry in docs/releases/v2.7.3.md and .ja.md	master (reaches next by the merge)	Ready to start
Subrequests: the content handler of a subrequest that returns Nginx::HTTP_OK without output sets status 500, where nginx’s return 200; gives 200. Decided on 2026-10-29 (option “500 のまま (推奨)”): 3.0.0 keeps the 500, so that auth_request fails closed (#574 keeps it). To do: the entry of #574 in docs/releases/v3.0.0.md and .ja.md already says that the status stays 500; add that nginx’s own return 200; gives 200 and that the 500 keeps auth_request from allowing the request	next	Ready to start
After a HEAD request to a location with an mruby output body filter, nginx sends the header and closes the keep-alive connection: the mruby body filter returns NGX_ERROR after it passed on the header of a header-only response (ngx_mrb_finalize_body_filter), and ngx_http_output_filter then sets c->error. The same with output, on 2.7.1, and on the merge of 2.7.2 into next (#628, which ran HEAD and then GET on one connection to such a location); docs/releases/v3.0.0.md (#574) names it as an exception (the review of #574, should fix 1). To do: a test commit (HEAD, then GET on one connection), then a fix commit	master (reaches next by the merge)	Ready to start
A rewrite handler that ends without Nginx.return (for example mruby_rewrite_handler_code 'x = 1' before a content handler, or before nginx’s return 200) sends no response, and error.log has no line at the error level: with keep-alive the connection stays open until keepalive_timeout, with Connection: close it is closed with nothing sent. ngx_mrb_finalize_rputs (src/http/ngx_http_mruby_core.c; the rc = NGX_OK after the log line “rputs_chain is null and return NGX_OK”) returns NGX_OK when nothing was sent and no status was set, and the rewrite phase of nginx 1.31.6 finalizes the request on NGX_OK (ngx_http_core_rewrite_phase), where the access phase goes on to the next handler. Found and measured on next and master while #583 was written and reviewed; no issue tracks it. To do: for v3, a test commit and a fix under the Pillar B row “Redesign response finalization per phase” of the plan, so that the request goes on to the next phase as with NGX_DECLINED; for 2.x, the behavior stays (both are the decisions of 2026-10-04), and a line under “Known limitations” in the release notes of the next 2.x release can document it	next	v3: waits for the row on step 4 below (the fix belongs to the Pillar B work that step 4 holds until the safety net is done); 2.x: nothing
An end-to-end benchmark scenario: throughput and p50 and p99 latency with oha or h2load, planned in AGENTS.md “Design principles” and in section 8 of the plan, which also records the reference proxy end to end	next	Ready to start; where the results are recorded is decided when it is added (AGENTS.md)
Site: steps 1–5 and the /v2/ implementation and workflows of step 6 are merged (#591, #607, #614, #617–#623, #626, #629–#632, #655, #657, #656). Step 7 deployment (#653) is live; see the verified deployment in the snapshot. #654 corrected the remaining master text after the switch, and its dispatched site build was verified live.	next for site/; master for 2.x sources	Cutover and master dispatch verified. Still open: the 3.x development banner of 3.4; the observability and certificate-lookup sample prerequisites of #619; the Japanese release-note line rewrites currently reported as NOTES_ONLY; and the owner’s go for Q9 (Discussions on, wiki off). New visual design is a separate follow-up, not part of this cutover.
nginx-tests with the modules that its skipped files need: a build with the options of test.sh and --with-http_v2_module, --with-http_v3_module, --with-stream_ssl_module, --with-mail and the other modules that “Results of nginx-tests” in docs/test/README.md lists (in the CI run of #602, 189 of the 209 skipped files need a module that test.sh does not build), with and without ngx_mruby, in the nginx-tests workflow	next	Ready to start: #602 is merged
nginx-tests with ngx_mruby code configured: runs that add handlers or filters with code for every request through TEST_NGINX_GLOBALS_HTTP (the Pillar E row of the plan), compared with the run without code. The pull request decides which code: a rewrite handler that ends without Nginx.return sends no response (the row on it above), so that handler would fail most files for a known reason	next	Ready to start: #602 is merged
Test harness rewrite (step 2 of section 6 of the plan; Pillar E): #635 (merged, c4e6ac6cc) made test.sh stop only the nginx that it started, by its pid, with no killall or pkill, and send SIGTERM again while the single nginx process still runs, because that process can miss a SIGTERM that arrives right before it waits for events (the run at 7926f72, where nginx stayed in epoll_wait() for 20 minutes after it logged the SIGTERM: “Stopping nginx” in docs/test/README.md has the cause and the measurements). #641 (merged, be559a017) added TEST_SH_PORT_OFFSET, which moves every port of the suite, so that two checkouts can run test.sh at the same time (“Ports” in docs/test/README.md). Still to do, in pull requests of their own: master_process on with several workers and reload cases, requests from CRuby instead of the client built into the test mruby, and HTTP/2 and HTTP/3 clients. A should fix of the review of #635 waits for the pull request that next changes test/build_release.sh, an input of the perf job: its comment still says that test.sh kills every running nginx. The readiness wait of #593 is the first part, on next since the merge of master at 6b27f08da	next	Ready to start: master_process on with several workers, on top of #641; no pull request yet
valgrind in CI: done on next by #634 (8bff5fbd7): valgrind errors fail the cell, one cell runs under valgrind on pull requests and pushes, and .github/workflows/valgrind-nightly.yml runs every cell except the sanitizer cell under valgrind nightly. GitHub runs a schedule only from the workflow file on the default branch, so the nightly run starts once a copy of the file is on master (“valgrind in CI” in docs/test/README.md). To do: a follow-up PR to master that adds valgrind-nightly.yml as it is on next; to run the cells of master under valgrind as well, the same PR gives the test.yml of master the input valgrind, the valgrind options, test/valgrind.supp and the step “Check the valgrind reports” of #634, and adds master to the matrix of the file	master	A decision that master, the 2.x line, takes this change of CI
nginx floor for v3 (item 2 of section 7 of the plan): drop the cells for nginx 1.26.3 and 1.28.3 from the build matrix of .github/workflows/test.yml and document the floor of 1.30 stable and 1.31 mainline	next	Ready to start: #601’s change to .github/workflows/test.yml is on next since the merge of master at 6b27f08da
OpenSSL 3.5 and 4.0 (item 3 of section 7 of the plan): cells for OpenSSL 3.5 as the baseline and 4.0 in .github/workflows/test.yml, which builds 1.1.1w today; drop 1.1.1 and 3.0, and check ENGINE and the removed APIs in ngx_mruby and the gems	next	Ready to start: #601’s change to .github/workflows/test.yml is on next since the merge of master at 6b27f08da
build_config.rb rewrite with explicit gems (step 3 of section 6; Pillar A): iij/mruby-env and iij/mruby-process replaced by core gems, the choice between core mruby-regexp and mruby-onig-regexp (item 1 of section 7 keeps onig-regexp for 3.0), monochromegane/mruby-secure-random and iij/mruby-digest replaced, and mruby-vedis and mruby-localmemcache out of the default build	next	Ready to start: #606, which edits the same file, is merged. The part that replaces env and process with the core gems needs the subtree at 4.1.0-rc2 (the row “mruby 4.1.0”); the other parts do not
mruby 4.1.0 (step 3 of section 6; item 1 of section 7). The step changes: the mruby subtree updated to the tag with Prism vendored (the tag’s tarball has an empty mrbgems/mruby-compiler/lib/prism, a submodule upstream), the three HAL gem names (hal-posix-dir, hal-posix-io, hal-posix-socket) removed from NGX_MRUBY_CORE_GEMS and its comment, the core env, process and regexp gems of 4.1 decided on (see the row above), build_config.rb.lock regenerated, the build with NGX_MRUBY_DEBUG_STATS run with its tests (4.1 keeps the roots in mrb->gc.root), the stream path of the counted GC registrations audited, a release notes entry, and the callgrind comparison against next, which includes the measurement of var and auth on mruby 4.1 (step 4 of the archived proposal for the method lookup of Nginx::Var). The owner decided on 2026-10-29 that next bundles 4.1.0-rc2 now and moves to 4.1.0 in a small subtree pull request when it is tagged	next	Nothing: the pull request for 4.1.0-rc2 is not opened yet
Build each bundled gem on mruby 4 (step 3 of section 6): #561 builds the default gem set on mruby 4.0.0; the rest is the build of all of them on 4.1.0-rc2 (the row “mruby 4.1.0”). auto-ssl is not built: the owner decided on 2026-10-29 to remove it from next	next	The pull request of the row “mruby 4.1.0” (not opened yet)
OpenSSF Best Practices badge (Pillar G of the plan)	master (README)	The owner registers the project on bestpractices.dev and gives the project id; then a session adds the badge
Step 4 of section 6 of the plan, the core stability work of Pillar B (execution context, response finalization, fiber lifecycle, common layer, headers and variables, and the two agent proxy rows), then v3.0.0-alpha.1	next	Held until the safety net of step 2 is done; what is left of it is the row above for the test harness rewrite (#641 is merged; master_process on with several workers comes next): valgrind in CI (#634) and the auto-ssl tests (#636 on master) are done. #600, #603 and #605 are merged (#603 on next since the merge of master at 6b27f08da, with the sanitizer log check of #601). The plan: nothing in the core is refactored before it exists
nginx-tests required: the nginx-tests workflow records its results and fails only when a build, the fetch of the suite or prove fails. In the runs of 2026-10-28 and 2026-10-29 the build with ngx_mruby and the build without it have the same counts (511 files, 302 passed, 0 failed, 209 skipped). To do: the job fails when the build with ngx_mruby fails a file that the build without it passes, and the job joins the needs of ci-ok	next	Ready to start: the counts are the same in every run since #602
The perf job in the needs of ci-ok, and the “In CI” text of docs/test/README.md (the sentences that call the job advisory and say that a FAIL does not block a merge, under “Performance comparison with callgrind”) aligned with the rule of AGENTS.md “Design principles”, under which a session does not merge while the job reports FAIL or ERROR	next, then master	Ready to start
Soak: record Pss and Private_Dirty from /proc/<pid>/smaps_rollup beside VmRSS, then bound their slope over more windows, set after measuring (the Pillar E row on memory accounting). Today test/soak/soak.rb bounds VmRSS by a step and a total and checks the counters of Nginx::Debug.stats	next	Ready to start
The ASan cell with -DNGX_DEBUG_PALLOC -DNGX_DEBUG_MALLOC (3.7 of the plan; the cell of test.yml sets only -fsanitize and -fno-omit-frame-pointer), so that ASan sees an access that runs from one pool allocation into the next. To do: a run of the suite with these flags, whose findings follow SECURITY.md, then the cell	next, master	Ready to start
A test build with ngx_otel_module (nginx/nginx-otel, built with CMake against the nginx source of test.sh) and an OTLP receiver for the case of the observability sample of 3.2 of the site proposal	next	The harness rewrite (the row above)
Step 7 of section 6 of the plan, the agent proxy, parts (a) to (e); none exists on next	next	Parts (a) and (c) wait, as step 4 does, for what is left of the safety net of step 2: the row above for the test harness rewrite (#641 is merged; valgrind in CI, #634, and the auto-ssl tests, #636 on master, are done); then they start beside step 4 (decision of 2026-10-29). (b) and (d) also wait for step 4. The order inside the step is a proposal of the 2026-10-04 amendment until the owner decides it
Remove example/ (README.md, nginx.conf, proxy.rb, set.rb, set2.rb, unified_hello.rb): no CI job starts it, no page links it, and its /proxy location picks among test1.example.com to test3.example.com, which are unreachable; examples/start/ is checked by examples.yml	next	Ready to start
When the handler of the location that Nginx.redirect runs sleeps and the handler that called Nginx.redirect then raises, the 500 page is followed by the output of the sleeping handler when its timer fires. The request ends and the connection is closed, and nothing leaks. Not a 2.x fix: it needs the rewrite of the row “The async resume rewrite” (#646 leaves it out)	next, in the core stabilization of 3.0.0	The row “The async resume rewrite”
The async resume rewrite (decision (c) of 2026-10-04, amended on 2026-10-29: the rewrite of how resumed handlers end in src/http/ngx_http_mruby_async.c, which includes the resume of a handler after Nginx::Async::HTTP.sub_request from its own request). No pull request exists	next, in the core stabilization of 3.0.0	Step 4, held as in the row on step 4 above
The upstream server setters: proposal. Step 1 is merged on master by #652 (8593d0d65): non-String TypeError, zone refusal, whole-name lookup, with English/Japanese 2.7.3 behavior entries. This integration carries it to next. The continuation request assigns options D/E to a separate v3 PR: ArgumentError for unusable String values and no synchronous hostname resolution in the HTTP setter, with API docs and English/Japanese 3.0.0 behavior entries.	next	Finish and validate this integration first. Reconcile the dated owner decision records before the v3 PR: the proposal still labels questions 2/3 open, and the v3 plan retains proposed agent-proxy ordering, release conditions and targets. Do not mark those unrelated proposals decided from the setter assignment. Keep TypeError, zone refusal, per-worker meaning and the existing IP/port/IPv6/unix distinctions.
The generated reference of the site (Pillar F of the plan): the directives from the ngx_command_t tables, the Ruby API with YARD and yard-mruby, the C API with Doxygen. site/scripts/check-reference.mjs checks, among other things, that each directive name has a heading in docs/directives/README.md; the pages are written by hand	next	Ready to start; no step of section 6 of the site proposal produces it yet

How a session continues

Use Continuing shared work for startup, ownership, checkpoints, temporary files and resumption. The rules are in AGENTS.md; this file records the queue and decisions, while each PR records its current head and evidence. The procedure applies to Claude, Codex and human contributors alike.

Private material

Vulnerability reports and their fixes are handled privately as SECURITY.md describes; this file does not record them.