Repository navigation
v0.9.63: integration validation fixes, mothership workflow fixes, Search MCP improvements - #10155
Conversation
…m llms.txt (#10140) * improvement(landing): add pricing faq and drop retired sdk claims from llms.txt * fix(landing): scope byok cost wording and use named heading size in pricing faq
…ws from Search MCP results (#10143)
…empties a date window (#10145)
…ettings surfaces (#10144)
…10146) * fix(lint): check table json inputs that hold bare block references Table input checks JSON-parsed raw field values, so a reference written outside a JSON string (e.g. {"count": <start.count>} or a whole-field <start.row>), which resolves to valid JSON at runtime, made the whole check end with an 'Invalid JSON' could-not-complete note. Quote those references in the lint-local copy before the Table block's own parser runs, so literal keys are still checked and the references reach the existing runtime-resolution notes. * fix(lint): defer table inputs whose quoted references fail predicate shape checks A quoted reference can still fail the Table block's shape validation, for example a filter in-list or a whole-field filter. Retry the transform without those runtime-only fields and note them as runtime-resolved, so the remaining literal inputs are still checked. Filter cases now run against the registered TableV2Block. * fix(lint): keep literal table input errors visible beside references Try string and one-item-list placeholders per reference instead of dropping any field that holds one, so in-list references are still checked and an invalid literal condition still reports. Only a field that is a single whole reference is deferred to runtime. * fix(lint): defer reference fields past the placeholder search bound Above the search bound an all-string or all-list fallback cannot tell a literal error from a reference type mismatch, so leave the reference-bearing fields out and keep checking the other inputs instead of ending the table check.
…ent window (#10147) * improvement(mothership): keep workflow test-run results inside the agent window * fix(mothership): measure the run result window after secret projection and fit the preview
…is reading (#10148) * fix(mothership): skip the 30s workflow pickup wait when no chat view is reading A client-routed workflow tool waited 30s for a chat view to pick it up before the server ran it, even when no view was reading the turn's stream at all (tab closed, user on another chat or page). Dispatch now checks whether the send's own response is still attached or a reconnected reader holds a short-lived Redis mark; with neither, it waits only 5s for a view that is re-attaching. A present view keeps the full grace, and an unknown (no Redis, failed read) counts as present. * fix(mothership): clear a reader's mark when it leaves and count it before its first event Each reconnect attachment now holds its own entry in a per-stream sorted set, removed on release, so leaving stops counting at once instead of lingering until the TTL. The route awaits the first mark before reading, and the Redis suite skips when no test Redis is configured locally.
…0149) * fix(lint): replace table placeholder search with structural placeholders Quote every bare reference, turn in/nin condition values into one-item lists and drop whole-condition references from all/any groups, then run the Table transform once. Only if that fails are whole-reference fields left out for one final attempt, whose error is reported. This removes the reference-count bound that skipped row data with many references, blames the literal error rather than a deferred field, and stops deferred row data from being called non-static. * fix(lint): track bare reference occurrences and defer reference-only filter groups Give each bare reference occurrence its own placeholder so a quoted value with the same text is left unchanged, rebuild objects with Object.fromEntries so a __proto__ row key survives, and treat a filter whose root group holds only references as runtime-only so the remaining inputs are still checked. * fix(lint): treat only single-key all/any objects as reference-only filter groups
…se long block runs (#10150)
…attaches during it (#10152) * fix(mothership): extend the workflow pickup grace for a view that re-attaches during it A reload drops the send's response, so a call dispatched right then got the 5s grace even though the view re-attaches about two seconds later. Presence is now read again when the short grace ends, and a view attached by then gets the rest of the full grace. A reader's refresh also drops lapsed entries, and the send's in-process attachment signal is covered by a test. * fix(mothership): never hold a finished workflow call on the grace extension check The presence re-check at the end of the short grace now races the browser's result, so a run that finishes or a Stop during a slow read returns at once, and the time the check took comes out of the extension.
…ry failed chunks three times (#10151) Presidio before 2.2.365 compared every match with every other one when removing duplicates and resolving overlaps, so a large text with thousands of matches took minutes. 2.2.366 replaces both with sorted sweeps. The mask client now makes three attempts per chunk instead of eight, since a retried chunk re-runs its whole NER pass.
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
There was a problem hiding this comment.
All reported issues were addressed across 51 files
Reply with feedback, questions, or to request a fix.
Turn on auto-fix | Re-trigger cubic
[High impact] The PR reaches shared Chat workflow execution, PII masking, search, and several user-facing settings and information surfaces. Fix the missing Computer Use controls before merging; the FAQ anchor is a smaller follow-up. FindingsSummaryUpdates workflow runs and chat reconnects, improves Search results and PII masking, exposes desktop settings on more pages, and refreshes public docs and pricing.
The new desktop settings pages hide Computer Use controls. The pricing FAQ also lacks its declared anchor.
|
… behaves (#10156) * fix(pii): describe the mask client's retry budget the way it actually behaves * fix(pii): name the exact retryable statuses in the retry budget comment
* docs(library): update best-ai-agents-for-it-helpdesk-teams * Pi Babysit: address PR #10159 feedback --------- Co-authored-by: Sim Pi Agent <pi@sim.ai>
…roval? (#10162) Co-authored-by: Sim Pi Agent <pi@sim.ai>
…s and IT (#10161) Co-authored-by: Sim Pi Agent <pi@sim.ai>
… perplexity, jina, duckduckgo, wikipedia and arxiv (#10179) * fix(integrations): validate exa, tavily, serper, linkup, parallel ai, perplexity, jina, duckduckgo, wikipedia and arxiv * fix(integrations): declare arxiv echo outputs, nullable result fields, and validate linkup max results * fix(arxiv): strip query and fragment from paper links and report the fielded search query * fix(linkup): treat an unset Max Results (null) as omitted * fix(integrations): send only Lite-supported Exa params, skip untouched Exa fields, widen the Tavily error reader, and clear stale Jina search values * chore(integrations): regenerate artifacts after rebase
* fix(integrations): validate postgresql, mysql, mongodb, neo4j and upstash * fix(integrations): stream neo4j records under the cap, keep mysql preferred tls as documented, tune truncation hints * fix(integrations): drop the row cap; stream mongodb and neo4j reads and fail at the response size limit
…dblue, linq, quartr (#10178) * fix(integrations): validate quiver, thrive, profound, agentphone, sendblue, linq, quartr - quiver: send retired Arrow 1.x model ids (arrow-preview, arrow-1.x) as arrow-2, offer Arrow 2 and Arrow 2 Telos in the model dropdown, cap max_output_tokens at the documented 65536 - thrive: get activity and get user by id call the documented /activities/{id} and /users/{id} paths - profound: mark raw logs and bot logs as retired by Profound (sunset 2026-06-10) - agentphone: mark the plan usage output as no longer returned by the API - sendblue: typing indicator status matches the documented SENT/ERROR values - agentphone, linq, thrive, profound, quartr: build single id path segments with safeUrlPathSegment * fix(agentphone): normalize numeric ids in response transformers * fix(integrations): keep retired Quiver models valid, skip Telos-rejected sampling fields, allow slashes in Thrive refs * fix(thrive): allow slashes in audience and manager references
* fix(integrations): validate spotify, buffer * fix(integrations): cap spotify playlist items limit at 50 and drop buffer tests for rejected output enums * fix(integrations): keep spotify get playlist tracks on the read endpoint that serves any public playlist * fix(integrations): batch spotify library requests up to each tool's old id limit, restore playlist tracks limit, list new buffer channel actions * fix(integrations): honor cancellation in later spotify library batches, report the exact rejected id range, reject unknown follow types * fix(integrations): reject spotify uris of another type, fail on invalid library check chunks, describe partial batch failures * chore(integrations): regenerate artifacts after rebase
…ter and temporal (#10181) * fix(integrations): validate polymarket, kalshi, sftp, ssh, smtp, jupyter and temporal * fix(integrations): honor negative position filters, type the new tests, and prove STARTTLS against a real SMTP server * fix(sftp): decode each escape run once and join upload names without decoding again * fix(sftp): reject a trailing parent segment in remote paths * fix(integrations): ignore null and empty timestamps in Kalshi settlements and Polymarket activity * fix(polymarket): reject a non-numeric activity timestamp and keep start=0 as full history * fix(polymarket): mark the next-page cursor output nullable * chore(integrations): regenerate artifacts after rebase
…password, persona, rootly (#10171) * fix(integrations): validate clerk, crowdstrike, infisical, ketch, onepassword, persona, rootly * fix(integrations): type the infisical approval fields as nullable and narrow the ketch test body * fix(integrations): return 400 for bad 1password path ids and reject unknown ketch consent values * fix(integrations): pass through unknown ketch consent values, show rootly title with detail, decrypt 1password vault details * fix(integrations): coerce rootly path ids once for the request body * chore(integrations): regenerate artifacts after rebase
…a, browser use, reducto, pulse and extend (#10180) * fix(integrations): validate firecrawl, context.dev, apify, bright data, browser use, reducto, pulse and extend * fix(integrations): fail cancelled polls cleanly and test at the HTTP boundary * fix(context-dev): send timeoutOpts only for a real timeout value * test(context-dev): check the untouched-timeout guard without a cast * fix(integrations): keep large Firecrawl results, cancel abandoned jobs, cap Pulse downloads * fix(firecrawl): cancel only a job Firecrawl returned an id for
…udits (#10183) * fix(integrations): close follow-ups from the integration validation audits * fix(integrations): type the polymarket activity test params * fix(integrations): stub fetch for every spotify library test * fix(integrations): migrate kalshi create and amend to the v2 order api, address review * fix(integrations): keep the kalshi book-side helper module-private * fix(integrations): route kalshi cancels by ticker and harden the read-first order writes * fix(integrations): keep every legacy kalshi order key on v1 write outputs
…rxiv link match (#10186)
…crawl job ids (#10187) * fix(integrations): keep saved output references valid and encode firecrawl job ids * fix(integrations): fail a firecrawl crawl or batch scrape that returned no job id * fix(integrations): keep open out of the public catalog and strip url control characters like the parser * fix(integrations): strip url control characters only where the id was sent raw
…efuses an advertised action (#10189)
…g ids, and trim zep final ids (#10190)
Uh oh!
There was an error while loading. Please reload this page.