github Skyvern-AI/skyvern v1.0.57
Release v1.0.57

48 minutes ago

What's Changed

Version bumped from 1.0.56 to 1.0.57

Recent Changes

  • Bump version to 1.0.57 (#9282) (c531db3)
  • Describe browser provisioning generically in two engine docstrings (SKY-20032) (#9287) (58fa8b7)
  • Accept a Power BI result that exactly fills the saved row limit (SKY-20044) (#9286) (aab78a4)
  • Give data-only Copilot code blocks a Goal that opens with their returned values (SKY-19371) (#9285) (96492c7)
  • Group Agent settings into six folding groups with change summaries (#9284) (c4e014f)
  • Start the display recorder bridge without OTel auto-instrumentation (SKY-19682) (#9283) (3dd728e)
  • Store Credentials-page secrets in AWS Secrets Manager on self-hosted AWS (#9281) (6c7d2ce)
  • Stop telling every deployment that credentials need a Bitwarden server (#9280) (90152d0)
  • Doc: shortcut hints in context, a grip beside each step, an insert line between steps (#9279) (e7e6135)
  • Reword three test comments about downstream readers (SKY-20031) (#9278) (c5910f3)
  • Copilot demonstration: keep a click that loads a new page (#9277) (b62114d)
  • fix(SKY-20024): list Demonstrate task after the matching steps in the Doc picker (#9276) (3961e83)
  • Start a task demonstration from the Doc's step menu (#9275) (3391b6d)
  • Doc: plain failure line, Enter runs the input box, truthful run results (#9274) (a0de4b5)
  • Make typing in the Doc behave like Notion: quiet new steps, Enter on empty makes a note, / renames or replaces (#9273) (88740f4)
  • Report start page and uploads on agent step nested-agent results (SKY-19823) (#9272) (d5f03a7)
  • Ask to add or move a login's site when Copilot's sign-in card would refuse it (SKY-19941) (#9271) (5e410cd)
  • Stop the API rescanning every module on each Playwright call from a frozen frame (SKY-18774) (#9270) (b549074)
  • Upgrade oauthlib to 4.0.1 in the cloud lockfile (SKY-20014) (#9269) (4b98af6)
  • Move the Nango services into one package with a single import surface (#9268) (731c53c)
  • Flag a Doc loop that goes through a whole step result instead of its list (SKY-19996) (#9267) (ffe6910)
  • Let users demonstrate one block of an existing agent after its earlier blocks run (SKY-19879) (#9266) (91493f4)
  • Release the save lock when the Doc's Run pre-save fails (SKY-19997) (#9265) (ccf3d21)
  • Fill a demonstrated credential in the iframe its field lives in (#9264) (2e2ee03)
  • Show what each Copilot test run cost and say why a test browser did not start (SKY-19921) (#9262) (473caf5)
  • Give each live view its own stream id and close a viewer whose relay stops (SKY-19931) (#9261) (0ba5e66)
  • Let Copilot take Word .docx SOPs from Upload SOP and the chat, and read them (SKY-19934) (#9260) (e6f31a5)
  • Stop saving Chrome's regenerated extension rules in browser profile archives (#9259) (f90df89)
  • Remove the ANALYTICS_DASHBOARD, onboarding_questionnaire_v1 and ENABLE_DISCOVER_COPILOT_HANDOFF flags now that they are on for everyone (SKY-4553) (#9258) (7248f4f)
  • Show a failed code block's own error, page and Fix with Copilot, and carry it in the API (SKY-19918) (#9257) (8ecd47d)
  • Wait for the browser to record a committed page before reloading it (SKY-19509) (#9256) (a75193a)
  • Keep the step you are typing in on screen after Cmd+S in the Doc (#9255) (d215d2b)
  • Count a tunnel failure as the site's when the proxy vendor could not reach the host (SKY-17599) (#9254) (027fe46)
  • feat: simplify Power BI and Gmail sign-in (SKY-19899) (#9253) (1800c72)
  • Document the receiver 503 and report owner_not_authorized for restricted credential_id triggers (SKY-18604) (#9252) (f1f938a)
  • Let code blocks read the text of a downloaded document (SKY-19726) (#9251) (86217c8)
  • Charge a code block's zero-action minimum once per label per run (SKY-19917) (#9250) (1be9cb2)
  • Turn off Edge component updates in Linux worker and PBS profiles (SKY-19857) (#9249) (5991e56)
  • Drop the internal doc link from the VNC keycode docstring (SKY-19952) (#9248) (950aaa2)
  • List an href-less link beside the Task V3 card whose click misses it (SKY-18877) (#9247) (08d206a)
  • Drop the global dependency override that rebuilt every request's dependency tree (SKY-18774) (#9246) (daf346d)
  • Make Doc steps read like Notion: no box on step text, controls on hover, chips on the text line (#9245) (c56274b)
  • Show the run-blocker list unless the Doc is actually on screen (#9244) (5983595)
  • Recover a statement the browser refuses to frame on a Task V3 download block (SKY-19922) (#9243) (441d830)
  • Keep a just-added Doc step quiet until you leave it, run or save (#9242) (92f1c17)
  • Quieter single-step runs in the Doc: no debug toast, no empty screenshot box, asks for missing inputs (#9241) (6639608)
  • Say a failed Doc step in plain words and offer Browser Task for multi-action Browser Actions (#9240) (86fca1e)
  • Place a visible CAPTCHA refusal on the visit whose document showed it (#9239) (cb02d0e)
  • Pipe the secret-mask test driver to node on stdin so CodeQL stops failing the OSS sync (SKY-14367) (#9238) (91c75f9)
  • Upgrade follow-redirects, body-parser and qs in the frontend lockfile (SKY-19561, SKY-19383) (#9198) (73fbad9)
  • Validate Bitwarden item ownership in Cloud (#9233) (4c948b2)
  • Keep prompt and provider text out of text prompt block logs (#9232) (08aba25)
  • Re-resolve a serialized run's saved profile at run start (SKY-19415) (#9231) (0024614)
  • Route Power BI query execution through Nango (#9229) (e9dabaf)
  • Run pre-commit and mypy as one 2-vCPU job, and check only the pushed range on main (SKY-19509) (#9228) (9ba3f1d)
  • Gate the Estuary capture-stall monitor on a slot that stopped confirming WAL (SKY-19625) (#9225) (cbcd7c8)
  • Run the CI pytest suite as 4 parallel shards (#9227) (57d1806)
  • Upgrade pypa/gh-action-pypi-publish to v1.14.3 in the SDK release workflow (SKY-19487) (#9184) (a203cab)
  • Upgrade qs to 6.16.1 and follow-redirects to 1.16.2 in the TypeScript SDK test lockfile (SKY-19244) (#9130) (139a954)
  • Upgrade zod to 4.6.7 and @hookform/resolvers to 5.9.2 in the frontend (SKY-18966) (#9095) (fdd426c)
  • Keep code-block action screenshots on responsive pages with slow PNG encodes (SKY-19607) (#9224) (5262f0a)
  • Upgrade serialize-javascript to 7.1.4 in the TypeScript SDK client pnpm lockfile (SKY-19426) (#9173) (636925b)
  • Keep the Stripe fence race test on one shared SQLite connection (#9223) (06f8809)
  • Run only the blocking Semgrep rules on PRs; move the registry-pack scan to a weekly and manual workflow (#9222) (92fb313)
  • Treat the ingress-off 404 as the dark state in the integration catalog (SKY-19523) (#9221) (5071938)
  • A2A agent: a status question in a conversation that holds no run asks which run (#9220) (4d3351e)
  • Hold a Task V3 type the same live dialog still refuses instead of retyping it (SKY-19408) (#9219) (b5681ee)
  • Skip the second sign-out read in the bearer role lookup (SKY-19051) (#9218) (4c938a1)
  • Offer an agent step login only from its own parameter_keys (SKY-19584) (#9217) (a7a75f4)
  • Stop a finished task's background mouse movement before its browser closes (SKY-19594) (#9216) (f0e5a72)
  • Enforce scoped API key permissions (#9215) (af5d1f7)
  • Stop captcha frame probes from hanging when the page navigates mid-read (SKY-19570) (#9213) (38d3380)
  • Let the Task V3 agent declare a job application in its first reply (off by default) (#9212) (27faf01)
  • Seed RBAC roles from live Clerk memberships (#9211) (a9c34c1)
  • Build the standalone tasks index test in a scratch database so an open snapshot elsewhere cannot hang it (SKY-19597) (#9210) (bed397d)
  • Run self-contained slow tests only when their inputs change (#9209) (a7f236a)
  • Record site-typed reCAPTCHA low-score rejections when the refused request is unambiguous (#9208) (b9ffcd0)
  • SKY-18494: Treat a blank upload setting as unset in the secret check and proxied S3 uploads (#9207) (1943673)
  • Show the browser Copilot is working in on the studio Browser pane (SKY-19459) (#9206) (db9a08a)
  • fix(studio): show a block screenshot on a step-index miss in the run view (SKY-19458) (#9205) (5d3f0ef)
  • Test the fork helper's timeout, exit-code and traceback paths (SKY-19532) (#9204) (b816a60)
  • Run the npm-audit runner pick on Blacksmith instead of GitHub-hosted (SKY-19484) (#9203) (bffbe24)
  • Add bounded Nango client and Power BI settings (#9202) (8ac4e0c)
  • Add Power BI Nango connection storage (#9201) (40b26bd)
  • Upgrade follow-redirects to 1.16.2 in the support agent lockfile (SKY-19561) (#9200) (3b11388)
  • Keep the integrations catalog loading when the forwarding availability check fails (SKY-19523) (#9199) (b8dbe98)
  • Record the application-context defaults ruling (SKY-18811) (#9196) (f86df26)
  • Add email forwarding setup and inbox UI (SKY-14152) (#9195) (314ea80)
  • Pin the select_combobox twin-row count and its unchanged-filter refusal (SKY-19549) (#9194) (d657191)
  • Merge an input[role=combobox] with its accessibility entry in observe (SKY-19544) (#9193) (e8c7292)
  • Clear an unverified secret field before failing closed (SKY-19543) (#9192) (4dde902)
  • Reject reuse of revoked email forwarding addresses (SKY-14150) (#9191) (6a415d0)
  • Receive and recover forwarded email 2FA codes (SKY-14151) (#9190) (eed9fa1)
  • Add OSS email 2FA forwarding ingestion foundation (SKY-14150) (#9189) (5d74abe)
  • Accept legacy Excel exports under .xls claims (#9188) (5023737)
  • Give an agent step with no tool list no tools, and start new editor steps with nine (SKY-19261) (#9185) (c7cd36d)
  • Normalize MCP workflow output summaries once instead of per nesting level (#9182) (5260ddd)
  • Wait for the blocked navigation's error page in the egress monitor browser test (SKY-19066) (#9181) (daeedc3)
  • Show the Browser pane in place of the Editor while demonstrating a task (SKY-19272) (#9180) (2720fe1)
  • Stop Copilot reporting every replayed demonstration step as failed (SKY-19358) (#9179) (940558b)
  • update worker config (#9178) (9d02837)
  • Wait for the pytest template to go idle instead of failing on a brief session (SKY-19086) (#9177) (48efb4a)
  • Create forced browser sessions only when the run starts (SKY-18937) (#9176) (dff1cd9)
  • Show why a credit top-up was refused on the billing page (SKY-19447) (#9175) (c9d37b1)
  • Doc editor slice 2: read-only Doc view behind WORKFLOW_DOC_EDITOR (SKY-18180) (#9170) (ff5b878)
  • Doc editor slice 1: notes and descriptions on workflow steps (SKY-18180) (#9169) (92bd563)
  • Bound workflow YAML block validation by dispatching on block_type (SKY-19396) (#9168) (d8e6ee8)
  • Match a failed run's closed-recording network facts against the user's error codes once (SKY-19084) (#9167) (fbc32a8)
  • Skip loading task extraction results when building run status (#9166) (80fa285)
  • Keep ad click IDs with marketing consent and store them with new accounts (SKY-18021) (#9165) (f94a1cd)
  • Add five SEO directories to the catalog as candidates (SKY-18055) (#9164) (0cd4567)
  • Let Task V3 type into a field that already holds the caret under an invisible layer (SKY-19320) (#9161) (b804c6e)
  • Cancel a trial by turning off the day-14 charge, keeping access to trial end (SKY-19338) (#9160) (7a30e0b)
  • Start the safety screen's backup provider when the first one hangs (SKY-18431) (#9159) (0bb9753)
  • Docs: sync with production (2026-10-28) (#9158) (123242a)
  • Make inline workflow secrets write-only (#9157) (a6e17ca)
  • Show signups by country group for each source in the marketing daily (SKY-19331) (#9156) (967fd75)
  • Load the saved browser profile when the review card's Test end-to-end runs (SKY-19307) (#9155) (fa2c00f)
  • Stop re-sending a failed run's screenshot to Copilot after a later run (SKY-19247) (#9154) (4c7249a)
  • Add the LinkedIn signup campaign by week to the Sales daily (SKY-18742) (#9153) (06127da)
  • Refuse block outcome writes from an execution that no longer owns the run (SKY-18224) (#9152) (4e144d4)
  • Keep a refused final write from shielding a stale run, and deliver external cancels to the running owner (SKY-18224) (#9151) (37dd6c4)
  • Keep a workflow run that won its terminal status write from being cancelled on lease loss (SKY-18224) (#9148) (97b82cb)
  • Let Copilot's rejected-login card offer Sign in myself and report outcomes as facts (SKY-19306) (#9147) (ebc8d62)
  • Save the pre-block export before Task V3's download baseline in its outside-hold test (SKY-18812) (#9146) (2e6d229)
  • Name the missing reference when a CodeBlock template fails to render (SKY-19251) (#9145) (3abea10)
  • Give saved-profile human-interaction runs a browser session the API can reach (SKY-18844) (#9144) (a57778d)
  • Copilot: stop rewiring conditional workflows, report skipped blocks, order parallel edits, show time left (SKY-19245) (#9143) (36b30b0)
  • Report anti-bot failures as system errors without a model call (SKY-19084) (#9140) (176dbde)
  • Update changelog — Week of October 26, 2026 (#9139) (c14028e)
  • Add Settings audit log page (SKY-18938) (#9138) (e633d4c)
  • Show the cost per paying customer for each ad platform in the marketing daily (SKY-17621) (#9137) (f90c553)
  • CASA 3.2.2 - Bind MCP OAuth redirect URIs to each registered client instead of a ChatGPT prefix (#9136) (1116b2a)
  • Restore three Ashby replacement reCAPTCHA tokens as a protective rollback (SKY-19087) (#9135) (c0efd0b)
  • Docs: sync with production (2026-10-27) (#9134) (d6a5255)
  • Email paying orgs that have run nothing for 56 days (off by default) (#9133) (53f2368)
  • Upgrade qs to 6.16.1 in the Claude Desktop bundle (SKY-19242) (#9132) (04e2c43)
  • Install the exact Node version the version file names in every image (SKY-19222) (#9129) (11ca4e2)
  • Reject invalid V1 browser-native downloads during finalization (#9128) (b58af52)
  • Keep a reaped run's final status when the lease check itself fails (SKY-18224) (#9127) (a217ce1)
  • SKY-19084 A2: own the final failure diagnostic only through this execution's conditional terminal write (#9126) (dbfd3a9)
  • Finish the reaper's task v2 branch through the workflow service (SKY-18224) (#9125) (aa83f63)
  • Refuse to start a child task once the run lease is lost (SKY-18224) (#9123) (bf79c97)
  • Sync the runs list inline when the orphaned run reaper fails a task (#9122) (ea376c9)
  • Pace run lease heartbeat retries after an exception (#9121) (3fbd655)
  • Keep the custom proxy credential test from failing on timestamp digits (SKY-18224) (#9119) (5211589)
  • Add agent step architecture docs, charter and routing skill (SKY-19145) (#9118) (7e23734)
  • Let authors add the agent block in the workflow editor when its flag is on (SKY-19135) (#9117) (11675d2)
  • Turn on posting for the Mon/Wed/Fri job application results report (SKY-19143) (#9116) (98b94e4)
  • CASA 4.1.3 - Encrypt stored secrets with AES-256-GCM, part 1: read support and decrypt-and-compare API-key lookup (#9115) (8c1652e)
  • Reject error pages on all download paths (#9114) (7a778ee)
  • Fall back to the live element handle when the driver cannot resolve a Task V3 act marker (#9113) (3623c43)
  • Send the card-required trial emails in production (turn off the dry run) (#9111) (d1b6258)
  • Fail complete-on-download blocks on partial success (#SKY-18526) (#9110) (236e61d)
  • Remove the PostHog HIPAA gate now that a BAA covers every PostHog product (#9109) (f629a2b)
  • Add an agent block that calls this run's tools until it can return schema fields, behind one flag (SKY-19057) (#9108) (a13d1d9)
  • Turn on trial auto-convert and dry-run trial emails in US production (#9107) (1ac8ab5)
  • Send GPT's earlier reasoning back with its tool calls (SKY-19191) (#9106) (ba7169d)
  • Check the hung rescue stage deadline without a wall-clock bound (SKY-19167) (#9105) (862fc81)
  • Keep pre-launch customers on Free when they leave a paid plan under the card-required trial (#9104) (a096b41)
  • Rebuild Patchright's selector scope tree when Chromium rejects a deep DOM describe (SKY-19055) (#9103) (77caed3)
  • Stop reporting the built-in missing-one-time-code error beside the block's own code (SKY-19162) (#9102) (0399724)
  • Keep a one-time code whole in a Task V3 field that drops typed keys (SKY-13943) (#9100) (dd76873)
  • Keep split-PDF page excerpts within the planner input token budget (SKY-19153) (#9099) (64493bc)
  • Replace Free plan copy with the 14-day card-required trial (#9097) (b04c3f7)
  • Restore LLMConfigRegistry after test_openrouter_error_propagation so the Copilot route test is order-independent (#9096) (49e948e)
  • Run Copilot on openai-agents 0.19.4 and openai 2.54 with no change in behavior (SKY-19058) (#9094) (46cb62b)
  • SKY-18537: Make cron-worker the only event-trigger dispatcher (#9091) (1b2cb4a)
  • Keep at least one vendor session worker running through node drains (SKY-19008) (#9084) (3caa168)
  • Upgrade vite, vitest and fast-uri in the TypeScript SDK client lockfiles (SKY-18764) (#8994) (0cf31d7)
  • Upgrade follow-redirects, vite and undici in the frontend lockfile (SKY-18764) (#8993) (7cbf78c)
  • Answer a rejected Bitwarden master password on the items listing with 400, not 502 (SKY-19014) (#9083) (81d4d07)
  • Build the step and task archive ZIPs on a thread so deflate does not block the event loop (#9080) (2094b01)
  • Hold a Task V3 run's first terminated verdict once and ask for the source of the stop (SKY-18980) (#9079) (5c4d7d2)
  • Look up a complete run or task ID by primary key in the runs list search (SKY-18408) (#9078) (46fcf74)
  • Keep the agent's evaluate calls in the page's main world (SKY-18988) (#9077) (7708569)
  • Mark the git blob SHA-1 in the copilot eval readiness test as not used for security (SKY-19002) (#9076) (c4022dd)
  • Fix the failing one-time-code browser tests in test_multi_field_totp.py (SKY-18990) (#9075) (0c28401)
  • Disable telemetry by default for self-hosted installs (#9073) (647b920)
  • Read files from an SFTP server as a File Download source (SKY-18299) (#9072) (23bd8af)
  • Read files from an SFTP server as a File Download source (SKY-18299) (#9071) (6329a81)
  • State the file count in download-block framing before the model decides (SKY-18769) (#9070) (ac83fc1)
  • Record impersonating staff in organization audit events (#9069) (d840fe7)
  • Right-align the self-hosted OIDC sign-out button (#9068) (733ca58)
  • Turn on the orphan-schedule sweep's backend deletes for the production cron worker (SKY-18564) (#9067) (9a5f750)
  • Pass the required arguments in the adopted-session blob download tests (SKY-18972) (#9066) (a228959)
  • Send Copilot tool schemas without generated titles on Claude and terra routes (SKY-18928) (#9065) (b8b7448)
  • Upgrade zod, undici and fast-uri in the internal dashboard lockfile (SKY-18965) (#9064) (017fceb)
  • Run finite warehouse curator dry runs locally (SKY-18921) (#9063) (7518239)
  • fix(ui): keep runtime API lookup valid (#9061) (c408eda)
  • Document round-one self-hosted identity QA results (#9060) (4bf9d1a)
  • Strengthen OIDC and RBAC regression coverage (#9059) (d944fbc)
  • On-prem admins can view signed-in members (#9058) (75fa163)
  • Audit OIDC sign-ins (#9057) (cb612b4)
  • Omit request bodies for restricted organizations (SKY-18935) (#9056) (7b0774b)
  • Stop logging workflow request and parameter values (SKY-18935) (#9055) (d9cf542)
  • Add OpenID Connect sign-in to the self-hosted UI (#9054) (b789378)
  • Add OIDC principals to self-hosted RBAC (#9053) (ce716f9)
  • Delete the Temporal schedules of deleted schedules in the orphan sweep, behind a flag (SKY-18564) (#9052) (98aeb9c)
  • Upgrade undici to 7.30.1 and body-parser to 1.20.9 in the Claude Desktop bundle (SKY-18930) (#9051) (77d4cc5)
  • Upgrade undici to 7.30.1 and vitest to 4.1.12 in the frontend lockfile (SKY-18931) (#9050) (af74f15)
  • RBAC v2 slice 4b part 1: frontend API layer for role settings (SKY-18909) (#9049) (863ed96)
  • RBAC v2 slice 6 part 1: detect the downgrade pause once (SKY-18910) (#9048) (ad49492)
  • Apply an agent's configured proxy region to API runs that omit one (SKY-18884) (#9047) (5dea641)
  • Save the page DOM on every Task V3 turn step of sampled runs, and flush each turn as it ends (SKY-18892) (#9046) (2455831)
  • Correct the drift-reader Phase 1 plan shape to what production plans today (SKY-14983) (#9044) (a7500d3)
  • Fix the Task V3 answer audit's quote check, dropdown and placeholder plumbing, and bound its holds per task (#9043) (c5bd690)
  • Keep the country calling code in an international phone field, and stop reporting a wrong one as entered (SKY-18890) (#9042) (c819046)
  • Record one step per Task V3 turn, with its model call and tool calls, without changing bills (SKY-18790) (#9041) (1b81905)
  • Return every in-loop 404/410 to the Task V3 model, behind a run-arm flag (SKY-16206) (#9040) (02e54f4)
  • Stop routing benchmark-flagged runs to the deleted Hetzner fleet and remove Hetzner code paths (SKY-18907) (#9039) (a903a05)
  • Keep Task V3 turn steps out of the per-step pricing skill's step counts (SKY-18896) (#9038) (7399bab)
  • Task V3 observe: name a Yes/No button pair's question on both buttons and echo it on click (SKY-18212) (#9037) (9e71516)
  • Charter: Task V3 gates act on recorded state, with a dated exceptions register (SKY-18888) (#9036) (abaf3df)
  • Assert the radio click gate agrees with the driver instead of naming one Playwright version's aria-disabled shapes (SKY-18878) (#9035) (7350bb5)
  • Drive the registered-browser deadline tests with a frozen test clock (SKY-18881) (#9034) (dd1f938)
  • Log which locator matches Copilot was shown as visible (SKY-18704) (#9033) (5eb5a79)
  • Upgrade pypdf to 6.21.2 and map its moved size-limit checks (SKY-18829) (#9032) (95e90db)
  • Let Copilot find and use logins from a connected Bitwarden vault (SKY-18784) (#9031) (c664b0a)
  • Save the final screenshot for Task V3 workflow blocks (SKY-18862) (#9030) (eccc1c6)
  • Let the Task V3 model select a replayed CSV instead of refusing it by shape (SKY-18752) (#9029) (76ab663)
  • Test the radio click gate against the driver's enabled check, not one Playwright version's rule (SKY-18878) (#9028) (daa3fce)
  • Align the real-Chromium tests with planned browser replacement, pinned Patchright, Playwright 1.56 and full-Chromium type-ahead (SKY-18752) (#9027) (c1e5a03)
  • Stop re-entering an externally supplied one-time code; generated TOTP unchanged (SKY-18619) (#9026) (b872da0)
  • Revert per-snapshot curator evidence admission (SKY-18507) (#9025) (b0cbc1c)
  • Persist run-end HAR encounters and page identity (SKY-18366) (#9024) (e6cfd45)
  • Upgrade Playwright and Patchright to 1.56.0 (#9023) (6cf38e4)
  • Fail the run at the first unexpected browser connection loss (SKY-18726) (#9022) (e9477af)
  • Upgrade graphql-core to 3.2.12 and nbconvert to 7.17.2 in the lockfile (SKY-18864) (#9021) (c7833a5)
  • Register a session browser before its first goto so later blocks see it (SKY-18860) (#9020) (5802141)
  • Log and alert when a saved workflow version adds a skyvern-1.0 engine pin (#9019) (e8f1028)
  • Serve browser test pages locally instead of from the internet (SKY-18854) (#9018) (637ea80)
  • Build a task run response when the task has no navigation goal (#9016) (e4be69c)
  • Add 19 agent benchmark cases built from failing production Task V3 runs (#9015) (9f08b9a)
  • A2A Agent: Implement A2UI cards (#9014) (8b57799)
  • Refuse browser egress to a destination we cannot resolve (SKY-18836) (#9013) (10b8d82)
  • Keep tests from downloading the public suffix and GuardDog package lists (SKY-18833) (#9012) (903d352)
  • Upgrade follow-redirects to 1.16.1 in the SDK test lockfile (SKY-18764) (#9007) (ff6a475)
  • Show Upload an SOP and Demonstrate task cards under the Home prompt box (SKY-18697) (#9011) (957faa9)
  • Handle interrupted native downloads in Task V3 (SKY-18624) (#9009) (218cb0c)
  • Label the anti-bot benchmark's API version as API service context with its sampling time (#9008) (0a786fe)
  • Treat an Accept 422 as proof the Copilot proposal was not saved (#9006) (5d657e1)
  • Count only the secure runner's own failures in the code block run-kill alert (SKY-18464) (#9005) (a59e8c4)
  • Restrict member temporary cleanup to test-login credentials (#9004) (0f23931)
  • Hold a v3 give-up once for the one permitted retry after a typed site rejection (SKY-17487) (#9003) (f94b4f6)
  • Document an email-forwarding path for email-based 2FA (#9002) (7927d1b)
  • Import the curator worker through its cloud package in the advisory boundary test (#9001) (98d0ef7)
  • Fix CodeQL polynomial ReDoS in MCP feedback report redaction (#SKY-18789) (#9000) (c1489cb)
  • SKY-18797: Retry event-trigger submissions that never reached Batch (#8999) (18b4a4d)
  • Run 12 pytest shards and 3 Vitest shards to cut PR CI wall time (#8998) (84f8e36)
  • Task V3: a code endpoint that answers without a code reads as "not yet", not an error (SKY-18390) (#8997) (5d647ec)
  • SKY-18798: enforce advisory-only curator authorization (#8996) (35bb602)
  • Require submit evidence (#8995) (6c486e7)
  • SKY-17536: Let MCP agents send feedback reports to the Skyvern team (#8991) (a3fe64c)
  • Bound the engine 1.0 element lookup so a frozen page cannot stall an action (#8990) (73e49ce)
  • Add final Gate 2 QA report (#8989) (273cef5)
  • Exercise late-loop diagnosis through the local Copilot agent loop (#8988) (d01583b)
  • Deliver the driver probe's late reply after the caller gives up, not on a timer (SKY-18759) (#8987) (03734a6)
  • Stop the PBS worker run-error test turning a slow run into TimeoutError (SKY-18760) (#8986) (3dcc567)
  • Grant warehouse capture access to challenge solves (SKY-18363) (#8984) (453c564)
  • Stop answer-tile clicks from being reported failed or toggled off (#8983) (63c6385)
  • Stop a self-hosted run from starting or writing loop output when another server holds its lease (SKY-18224) (#8982) (00ad00f)
  • Protect API-key creation and provision role fixture key (#8981) (9e0691c)
  • fix: block Zillow first-party PX bootstrap (SKY-18052) (#8980) (5ba0c70)
  • Default new navigation blocks to no model override and drop empty model names on save (#8979) (016df8c)
  • Hold the leave dialog's Save and Copilot Save inputs while authoring owns the editor (SKY-17997) (#8978) (2d761e7)
  • Let Copilot offer a one-click demonstration when the user says they'll show it (SKY-18725) (#8977) (fd35bea)
  • Record element-to-element drags in demonstrations and add a Copilot drag tool (SKY-18727) (#8976) (7f47c20)
  • Keep runner stalls from expiring broker test requests before they are forwarded (SKY-18688) (#8974) (3c758c9)
  • Parameterize Copilot eval models and report driving costs (#8973) (132c84e)
  • SKY-18729 Demonstrate task: keep a skipped sensitive-field step in the step list (#8972) (7f0edf3)
  • Retry a refused LLM call once on the route's fallback, then fail it (SKY-18342) (#8971) (6a1de0c)
  • Show Copilot the recorded error of a page call a completed block caught (SKY-18206) (#8970) (0fc4361)
  • [HOLD] Prepare disabled curator production wiring (SKY-18657) (#8969) (926f472)
  • SKY-18718 Demonstrate task card: clearer description copy (#8968) (e729b3f)
  • Hide Edit for imported 1Password credentials (#8964) (079351c)
  • Update API specifications with fern api update (#8904) (623eb57)
  • Add curator rollout and inherited consent controls (SKY-18570) (#8963) (a4fe637)
  • Demonstrate task: consistent naming and entry point (SKY-17916) (#8962) (2910995)
  • Disable Save while an authoring action or Copilot change holds the editor (SKY-17997) (#8961) (2b9edf9)
  • Run every Task V3 test through production's agent setup (SKY-18395) (#8959) (1822679)
  • Copilot: size the history budget from each model's pricing and context window (SKY-18319) (#8958) (dee4a47)
  • Wait for the test child to finish exec before the browser-process check (SKY-18575) (#8957) (0e75b6a)
  • Copilot: keep the current screenshot when the token budget's last-resort prune runs (SKY-18319) (#8956) (edafdaa)
  • Stop the older Copilot turn when a new turn starts in the same chat (SKY-18319) (#8955) (29271c4)
  • Add 1Password credential import UI (#8953) (4e681d0)
  • Import and refresh selected 1Password items (#8952) (e604ee3)
  • Complete imported 1Password credential surface parity (#8951) (34400ad)
  • Restrict access to credential folders (#8950) (1ae950b)
  • Keep credentials and request bodies off cross-origin redirects, and internal errors out of workflow-create responses (SKY-12526) (#8949) (2e02161)
  • Make MCP sign-ins use member-bound API keys (#8947) (45dc4a0)
  • Classify job-recipe bot checks separately from login failures (SKY-18228) (#8946) (2d274fc)
  • Put a space between monitor handles that ran together so their alerts reach Linear and Slack (SKY-18474) (#8945) (599e656)
  • Raise the production 1vcpu patchright fairness lane to 450 pods and lower the 0.5-vCPU lane to 350 (SKY-17804) (#8944) (6468aa8)
  • Remove two no-action monitors and replace the any-5xx alert with a per-route 5xx count (SKY-18419) (#8943) (9f30f1e)
  • Arm the reconnect tests' 1 s guard once the stop under test starts (SKY-18498) (#8942) (0841e73)
  • Mirror the billing ledger tables to the warehouse and declare the Stripe mirror as a dbt source (#8940) (d5d64c2)
  • Fix the growth-funnel unit-test fixture so the nightly dbt build runs again (SKY-17880) (#8939) (98045a1)
  • Add event trigger dispatcher (#SKY-18483) (#8938) (efa027d)
  • Harden script sandbox validation (SKY-18441) (#8937) (31a64c1)
  • SKY-18462 Add signed event trigger receiver (#8936) (0dfbd6c)
  • Apply the merged-but-unapplied Datadog monitor definitions, and make route-5xx applyable (SKY-15449) (#8935) (8604b8d)
  • Stop closing the Studio editor from marking a workflow unsaved (#8934) (88b7115)
  • Add cloud event trigger management API (SKY-18463) (#8933) (1e5e185)
  • Remove the credential-rejection confirmation gate (SKY-18359) (#8932) (ab88b56)
  • Add trigger receipt tables and repositories (SKY-18377) (#8931) (da5a164)
  • Delete the two captcha wall-clock tests that turned main red (SKY-18442) (#8930) (74ef06f)
  • Switch on Flutter web accessibility semantics so both readers can see canvas-drawn apps (SKY-18263) (#8928) (d5e79eb)
  • Give both concurrent-index migration tests a scratch database so an open transaction cannot stall the build (SKY-18402) (#8926) (e66c233)
  • Fail runs orphaned by a crashed self-hosted server (SKY-7819) (#8925) (8e73b4d)
  • feat: attribute builder integrations in run telemetry (#8924) (3b67204)
  • Send Claude effort keys' effort on Task V1 and cache its repeated task text (SKY-18372) (#8923) (b6c6fae)
  • Upgrade mcp to 1.30.0 in the lockfile (SKY-18400) (#8922) (bed2561)
  • Merge the retention pod weekly report into the daily report (SKY-18314) (#8921) (e4a9145)
  • Skip DataDragon ledger DDL when the schema is current so concurrent runs stop deadlocking (SKY-18309) (#8920) (010331f)
  • prevent HIPAA organizations' browsers from sending anything to PostHog (SKY-18315) (#8913) (48ff7d7)
  • Pin claude-code-action to v1.0.165 in the Claude workflows (SKY-18245) (#8916) (ee70e25)
  • Upgrade ajv to 8.20 in the skyvern-ts client pnpm lockfile (#8914) (f6d0413)
  • Docs: add Grok Build to MCP setup (#8912) (3d389a0)
  • Record run.completed, plan on the organization, and upgrade starts on the portal route (SKY-18287) (#8911) (acbce16)
  • Stop HIPAA organizations from sending anything to PostHog from server code (SKY-18286) (#8910) (268421b)
  • Keep each run's LLM choices in its own context instead of the shared app (SKY-16970) (#8908) (7566558)
  • Read organization_pricing.ended_at in the staged pricing model so ended contracts stop accruing revenue (SKY-18295) (#8907) (fa85868)
  • Give a tokenless Copilot Accept an outcome the client can read back (SKY-17870) (#8906) (717519a)

Full Changelog: v1.0.56...v1.0.57

What's Changed

Full Changelog: v1.0.56...v1.0.57

Don't miss a new skyvern release

NewReleases is sending notifications on new releases.