Repository navigation
fix: backport eight user-reported fixes to 0.12.1 (batch 22) - #31987
Conversation
…#31346) The vector store's lesson ranking, findings admission and per-turn lessons split text on \w+, which keeps an unspaced Chinese sentence as one token, so a Chinese request shared nothing with a Chinese lesson on the default memory tier and its findings stayed withheld. Those read paths now add, on the request and the lesson side, the non-CJK pieces and two-character pairs split_cjk_word already gives the JSONL path. Latin text tokenizes as before, and lesson_keywords (the dedup's overlap) still keeps a CJK run whole. (cherry picked from commit c6251c0)
#31390) A cron run that lost some, not all, of its tool calls to the security gate shows "Partial" in the Schedule page's job list, but the Executions tab and the job's Logs tab showed the same run as "Error". The history keeps status "failure" for it on purpose, so both views now read the refusal from the run's error, the same reason the job row reads, and show the same Partial badge. Closes #24616 (cherry picked from commit 81d3fe3)
) The edit-person dialog could drop a chosen picture when Save landed in the instant the picture read finished. Two stale views caused it: - The save mutation read the name and picture from its render, and useMutation hands a new mutationFn to its observer only in a passive effect. A Save between the commit that enables it and that effect sent the form from before the picture. - An Enter in the name field after the read's promise ended but before React rendered it ran the old submit, which saw the read as done and saved without the picture. The form now rides on mutate(), and a submit from a render that still shows the read is ignored. The first is the race behind the intermittent PeoplePanel test failure on a loaded shard. Closes #26862 (cherry picked from commit 58a1213)
) Adding or editing a semantic entry on the Vector Memory card, and the global fact form in the memory editor, showed the English error prose of PUT /api/memory/semantic for its coded refusals. owner_only and store_unavailable now read from the catalog in every language; any other refusal keeps its prose. Closes #25592 (cherry picked from commit 38c0dc5)
Room tone left after a pause in a streaming dictation is loud enough to pass the digital-silence check, so the trailing utterance is decoded and whisper turns it into a lone "you" or "Thank you", which lands in the composer and, with submit-on-release, goes straight to the agent. A later utterance whose detector never latched speech and whose whole final is one of those two fillers now yields an empty final. Any other text is kept, because a short word after a noise-gated pause can also fail to latch the detector; a session's first utterance and a filler the detector heard spoken are kept too. Closes #29279 (cherry picked from commit cd8f6d9)
…31531) Papyrus accepted a project name ending in '.' and a Windows device name such as 'nul'. On Windows the first is created, listed and opened under another name ('draft.' becomes 'draft'), and the second answers 409 "project already exists" because the device exists in every folder. Create and clone now refuse both on every platform, before the "already exists" check, with a JSON 400 coded unportable_project_name, using the shared WINDOWS_DEVICE_STEMS set. Opening an existing project is unchanged. Closes #27868 (cherry picked from commit bf9e89e)
…1541) On a platform where the webapp preview's local copy is not served (anything but Linux and macOS), the mint endpoint answered {"available": false} without remote_framable, so the card lost the live-deployment iframe as well as the local preview. The platform gate now makes the mint a regular miss after the direct-local check, so it carries remote_framable and writes the same denied audit as every other miss, and a caller that is not direct-local still triggers no framability probe. The local file route stays off on those platforms. (cherry picked from commit 20c71e2)
Three diagnostics matched an SSH login refusal only as the fixed text "Permission denied (publickey". OpenSSH prints the server's auth methods in the server's own order, so a server answering "Permission denied (keyboard-interactive,publickey)." got none of them: the notes app dropped the cause line from its git error, the prepare-pr push probe gave the fork remedy instead of the credential one, and the base fetch hint said "permission denied" instead of "run ssh-add". Each now accepts any parenthesized, comma-separated list of method names closed by ")", the rule the app-install clone classifier got in #26371. A local file error's bare "Permission denied" carries no such list, so it still does not match. The two prepare-pr scripts share one pattern, since preflight already imports from push_guard. Closes #26367 (cherry picked from commit 07ab5da)
First Principles Review (Opus 5.5) — ✅ PASSPremise-level review of First-Principles-Verdict: PASS This backports eight fixes that are already merged on main. Each one has a linked user report, and none adds surface beyond its own fix. What this change shipsInventory (8 items) — 8 justifiedIntent: ship eight user-reported fixes from
[FIRST-PRINCIPLES-REVIEWED] 849e2f0 Description read[DESCRIPTION-READ] d6049271839d603f619205aed59ae8a9675257c832400df7c30f0e0eb2dc9a6e That sha256 names the PR description this verdict read. Recompute it with |
UX Review (Opus 5.5) — ✅ PASS (a BLOCK with no Trigger and no Outcome does not block)UX-level review of UX-Verdict: PASS cannot evaluate: missing screenshots of the "Partial" badge (Executions and Logs) and the two new Vector Memory error messages. The PR attaches no image, so the blind read was not performed. Evidence gaps
Notes
[UX-REVIEWED] 849e2f0 Description read[DESCRIPTION-READ] f6e7ba846008899a490b8cc2b470530964dac738b52eae7bdeebb965e89c1979 That sha256 names everything this verdict read: the PR description, plus the 3 evidence file(s) the workflow collected from it and handed this review. It covers that evidence because the media strip replaces every attachment URL with the same placeholder, so an edit swapping one attachment for another would leave the description's captured text identical while what this review judged changed. Recompute it with |
Design Review (Opus 5.5) — ✅ PASSDesign-level review of Design-Verdict: PASS Nothing blocks: these are eight clean cherry-picks of fixes already merged on Notes
[DESIGN-REVIEWED] 849e2f0 Description read[DESCRIPTION-READ] 1205b103909621a723a12243c1c630328f326a26a330b917298ac83715f1fbb4 That sha256 names everything this verdict read: the PR description, plus the 1 evidence file(s) the workflow collected from it and handed this review. It covers that evidence because the media strip replaces every attachment URL with the same placeholder, so an edit swapping one attachment for another would leave the description's captured text identical while what this review judged changed. Recompute it with |
Opus 5.5 Review — ✅ no blocking findingsReviewed Verdict parsed from the review's SHA-scoped output markers for commit False positive or not applicable? A repository writer can comment: |
GPT 6.1 Review — ✅ human override acceptedHuman judgment by @buluoray overrides the GPT 6.1 finding for [GPT-OVERRIDE] 849e2f0 This comment is updated in place on each push. The model was not re-run because an authorized human decision supersedes it. False positive or not applicable? A repository writer can comment: |
Human judgment recorded@buluoray recorded a not-applicable release (class
This decision applies only to this commit. A new push requires a new judgment. |
Problem / Motivation
Goal: Bring eight small fixes for user-reported issues that merged on
mainafterv0.12.0ontorelease/0.12.0, one commit per fix, so the 0.12.1 patch ships them and each one can be reverted alone.Why it matters
draft.,nul).Not a goal
main; nothing is changed only here.CHANGELOG.mdchange. Cuttingv0.12.1-rc.Nand the bare tag follow docs/build/release.md and stay with the maintainers.What changed (motivation → approach → change)
Each main squash commit was cherry-picked with
-xontorelease/0.12.0@84a49d097b, one commit each, keeping its original author. The order is merge order except #31329, which was applied last; it touches no file the other picks touch.c6251c02bc2e14c5403b81d3fe3e602029d71a2758a121336f08aa85191d38c0dc5e1e1548e28b26cd8f6d90050ed028e372bf9e89ed0cf9a63669b420c71e235ee2d4f71b7807ab5da194849e2f0769No hunk was resolved by hand. Every pick's added and removed lines are identical to its main commit. #31531 and #31329 edit the built-in Papyrus and md_notebook apps in place, which AGENTS.md allows until they move out. #31329 changes how an SSH error message is matched for diagnostics; it changes no credential or login handling.
builder-security checked: #31541 edits the
webapp_previewhandler. It changes no bind, route or auth check. On platforms other than Linux and macOS the local file route still serves nothing; the mint now answers a regular miss there, carryingremote_framableand writing the same denied audit as every other miss, and a caller that is not direct-local still triggers no framability probe. No other pick changes a listener, an ingress rule or a storage grant.Backwards compatibility
Compatible: each change is the one already merged on
main, and each PR's own compatibility note applies unchanged. Papyrus refuses a project name ending in.or naming a Windows device.Removes nothing: no pick hides, deletes or renames anything 0.12.0 accepts.
Tests
Proof: not provable -- other: a backport of main commits whose proofs ran on their own PRs
release/*, so the related set was run here: every touched test file plus every test file that imports a touched module, 21 files, one pytest process per batch of 20,-n 0. 1289 passed, none failed.src/i18n/*.test.ts*catalog test, 59 files, 1179 passed.Manual verification
N/A - every pick was verified on its own main PR. The 0.12.1 RC soak is the end-to-end check.
Screenshots / video
Why no screenshot: each frontend change here is applied from
mainunchanged; its look is reviewed on its own PR, not here.Related Issues
Backport to the 0.12 line for 0.12.1 of #31346, #31390, #31282, #31347, #31515, #31531, #31541 and #31329. Source issues: #25649, #24616, #26862, #25592, #29279, #27868, #27836, #26367.
Sibling sweep
Siblings: none -- each source PR carried its own sibling sweep; this PR only carries those fixes.
Pattern harvest
Not generalizable: this PR only carries fixes already on
main. Each source PR has its own harvest.Checklist
feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)Contribution License Agreement
Contributions are accepted under the project's Apache-2.0 license; see the Licensing section of CONTRIBUTING.md.