Repository navigation
v0.9.60: pii tokenizer fix, gitlab group sources, computer-use reliability - #10114
Conversation
…an't block mobile taps (#10101)
…fuses as unsupported, with the snapshot still valid (#10106)
…he execution snapshot (#10104) * fix(resume): stop shipping the execution snapshot to the resume page * improvement(resume): capture pause block names at pause time and keep display reads off the snapshot
* docs(readme): match integration counts to the generated catalog totals * docs(rules): use the catalog-derived integration count in brand rules and docs manifest
…tised action as unsupported (#10109) * fix(computer-use): treat an attribute-unsupported refusal of an advertised action as unsupported * scope the refusal mapping to performed actions; value and selection writes keep the generic error
…10107) * feat(search): add GitLab group sources with per-project permissions * fix(search): preserve GitLab group scope across connection settings * fix(search): validate current GitLab group access and preserve auth errors * fix(search): preserve connector setup contracts when changing scope
…missing health pings (#10111) * fix(pii): tokenize long unbroken runs in linear time so workers stop missing health pings spaCy's tokenizer strips a whitespace-free run one prefix or suffix at a time and re-searches the whole run for each, so a long run such as minified JSON costs quadratic time with the GIL held. A 100k-character NER window of minified JSON held it for seconds, the worker missed uvicorn's 5s health ping, and the supervisor killed it along with every request it was serving. Wrap each model's tokenizer so runs of 100+ non-space characters split in linear time into words and single symbols, leaving doc.text and all other tokenization unchanged. Add an e2e check that fails on the old tokenizer. * test(pii): check PII on both sides of the long run is masked in place
…after launching it (#10112) * improvement(computer-use): wait for an app's first observable window after launching it * count a minimized window as ready and time out each window read
…log (#10113) * improvement(library): hold sim's integration, action, and trigger counts to the catalog in check:library-content * fix(library): match bold and title-cased sim counts, and track code fences by length
|
The latest updates on your projects. Learn more about Vercel for GitHub. |
[High impact] GitLab group search changes who can find existing project content, while other updates affect bounded workflows and published guides. The code changes appear safe, but the new mock-only test must meet the repository's testing requirement before merging. Findings
SummaryThis release adds GitLab group sources, speeds up PII tokenization, improves desktop computer use, and keeps large execution snapshots off the resume page.
|
There was a problem hiding this comment.
All reported issues were addressed across 77 files
Reply with feedback, questions, or to request a fix.
Turn on auto-fix | Re-trigger cubic
…estors when it reports none (#10116)
|
@cubic-dev-ai review this PR |
@waleedlatif1 I have started the AI code review. It will take a few minutes to complete. |
There was a problem hiding this comment.
All reported issues were addressed across 77 files
Reply with feedback, questions, or to request a fix.
Turn on auto-fix | Re-trigger cubic
…alized words (#10115) * fix(library): attribute sim counts by clause and table row, not capitalized words * fix(library): handle colon lists, parenthetical comparisons, relative clauses, prefixes, and linked names in sim-count * fix(library): keep sim relative clauses attached, anchor parenthetical comparisons to sim, and detect pipe-less table rows * fix(library): remove sim-count comparison asides for any subject and keep counted openers * fix(library): keep sim-count comparison spans inside their sentence * refactor(library): read sim-count comparisons as their own clauses instead of deleting them * fix(library): split sim-count contrasts nested in comparisons, keep count lists together, and end sentences before closing quotes * fix(library): start sim-count clauses only at a new capitalized subject and tell count lists from new clauses * fix(library): check sim claims in feature-first table cells and treat digit product names as new subjects * docs(library): describe sim-count table cells and new-subject clauses accurately
…10118) The SSE and NDJSON line readers appended each decoded chunk to the pending line and re-split the whole buffer, so one large event (a block output of a few MB in the run stream) was rescanned and re-flattened on every chunk: quadratic in the event size. The pending remainder never holds a newline, so splitting just the new text and prefixing the first piece gives the same lines.
Uh oh!
There was an error while loading. Please reload this page.