Changelog

Improvements

• Copilot Live Status Narration — The workflow editor copilot now displays a short live status line while it works, so you can see what the agent is doing instead of waiting in silence. (#5567)
• Renamed Browser Profile Reset Action — The API endpoint for clearing a persisted browser session has been renamed from /browser_session/refresh to /browser_session/reset_profile for clarity. The old path remains available as a deprecated alias. (#5572)
• Fewer Screenshots for API Runs — Screenshot capture frequency has been reduced for API-triggered runs, lowering storage overhead on long-running automations without affecting interactive sessions. (#5565)

Bug fixes

• Fixed the workflow copilot incorrectly proposing a workflow it had just failed to build. The copilot now only surfaces an Accept/Reject panel when the last test actually passed. (#5574)
• Fixed 2FA entry failing on pages with custom overlay input components where the real input field is hidden. The agent now types digit keys directly when no interactable input exists. (#5568)

New features

• Saudi Arabia Proxy Location — Added RESIDENTIAL_SA as a proxy geolocation option for automations targeting Saudi Arabian websites.
• Workflow-Level Error Code Mapping — Workflows now support an error_code_mapping field that all blocks inherit automatically. Block-level mappings still take precedence, giving you a single place to define default error handling without repeating it on every block. (#5506)
• AI Summarize for Workflow Run Outputs — A magic-wand “Summarize with AI” button now appears next to block and workflow run output sections, generating a plain-language summary of raw JSON output on demand. (#5556)
• Reset Browser Profile API — A new POST /workflows/{workflow_permanent_id}/browser_session/reset_profile endpoint clears a corrupted persisted browser session so the next run starts from a fresh browser state. (#5544)
• MCP OAuth for Remote Clients — MCP clients (e.g. Claude Code, Codex) can now complete browser-based OAuth to authenticate with Skyvern without getting stuck in redirect chains. (#5558)
• OpenClaw MCP Setup — skyvern setup openclaw and skyvern mcp switch now support OpenClaw as an MCP client target for both remote and local stdio configurations. (#5536)

Improvements

• Smarter Extraction Cache Keys — The extraction cache now normalizes session-volatile tokens (random UUIDs, CSRF values, nonce query params, timestamps) before hashing cache keys. Semantically identical page visits now share cached extraction results, improving cache hit rates across workflow re-runs. (#5504)
• Reduced Rate-Limit Errors on Extraction Prompts — Extraction prompts are now capped in size before being sent to the LLM. Oversized previous_extracted_information, extracted_information_schema, and element trees are truncated rather than sent whole, significantly reducing 429 RESOURCE_EXHAUSTED errors on high-volume workflows. (#5502)
• Cross-Run Extraction Cache — Extraction results are now cached in Redis across workflow runs, scoped to the workflow. Recurring scheduled workflows skip redundant LLM calls when the page content hasn’t changed. (#5562)
• Copilot Survives Tab Switches and Reconnects — The workflow copilot agent now runs to completion even if the browser tab is closed or the connection drops mid-request, and the AI reply is persisted so it appears when you reconnect. (#5560)
• Copilot Fails Fast on Unresolvable URLs — The workflow copilot now immediately surfaces the real error (e.g. DNS resolution failure) and stops retrying when a URL cannot be reached, rather than reporting success. (#5563)

Bug fixes

• Fixed CSV parsing failing on files whose first header row exceeded ~1 MB, causing valid wide-format CSVs to be rejected with “Could not determine delimiter”. (#5510)
• Fixed a browser automation crash on sites with cross-origin stylesheets (e.g. MUI-based apps). CSS hover styles are now parsed entirely in memory without modifying the DOM, preventing React reconciliation errors that caused the agent to loop until max steps. (#5508)
• Fixed context-window overflow failures being misdiagnosed. Steps that fail due to oversized input are now correctly counted and short-circuit to a clear LLM_ERROR failure reason instead of triggering a redundant LLM-based failure analysis. (#5507)
• Fixed workflow file-download blocks losing their configured download_suffix at task finalization, causing downloaded files to be persisted without the expected filename suffix. (#5499)
• Fixed cached scripts for non-ATS workflows incorrectly generating fill_form() calls, which defeated the performance benefits of caching by delegating back to AI at runtime. (#5497)
• Fixed the pagination navigator on the Workflow Past Runs and Run History pages being right-aligned instead of centered.
• Fixed run recordings missing an end timestamp and broken scrubbing. Recordings are now finalized with a stream-copy remux so players correctly show duration and support seeking. (#5551)
• Fixed tag input fields (e.g. skill set fields on job application forms) not selecting values from the dropdown. These fields are now correctly treated as auto-completion inputs. (#5557)
• Fixed per-step video sync spawning a redundant ffmpeg process on every step, causing CPU spikes and log noise. (#5559)
• Fixed workflow parameter edit and delete buttons being hidden when a parameter name was too long.
• Fixed webhook delivery failing silently when a stored webhook URL had a leading or trailing space. (#5550)

New features

• Adaptive Caching (Script Generation) — Skyvern can now learn from repeated workflow runs and generate cached scripts that speed up future executions. When a block runs successfully, Skyvern records a reusable script and replays it on subsequent runs, falling back to the AI agent only if the script fails. (#4908, #4916, #4917, #4920, #4922, #4931)
• Workflow Trigger Block — A new block type that lets one workflow trigger another from within the workflow editor. Chain workflows together as building blocks for complex automations. (#4885)
• Browser-based CLI Signup — New users can sign up for Skyvern Cloud directly from the CLI. Running skyvern signup opens a browser flow that creates your account and stores your API key locally. (#4925)
• Interactive ngrok Tunnel — skyvern browser serve can now create an ngrok tunnel automatically, making it easy to expose your local browser to Skyvern Cloud without manual network configuration. (#4924)
• South Korea Proxy Location — Added RESIDENTIAL_KR as a proxy geolocation option for automations targeting South Korean websites. (#4918)
• Downloaded Files Tab — Browser session detail pages now include a Downloaded Files tab for viewing and accessing files captured during a session. (#4911)
• CDP Proxy Authentication — Remote browser connections now support authenticated proxies via the CDP Fetch.authRequired protocol. (#4936)
• Remote Browser Download Interception — File downloads triggered by XHR requests and other non-navigation events are now captured when using remote CDP browser connections. (#4906, #4921, #4934)

Improvements

• Renamed “Login-Free” to “Saved Profile” — Clearer terminology throughout the UI. Browser session checkbox wording has also been improved. (#4914)
• TOTP Verification Improvements — Better timer handling, expired code retry, 2FA banner suppression, and more reliable goal-text extraction during TOTP flows. (#4860)
• TOTP Webhook includes workflow_permanent_id — Easier to identify which workflow is requesting a 2FA code. (#4871)
• Consolidated Gemini 3 Pro Model Key — VERTEX_GEMINI_3_PRO now auto-resolves to the latest version, so you no longer need to update config when Google releases point versions. (#4926)
• Browser Rotation for Remote Connections — Remote browser environments now support context rotation, improving reliability for long-running automations. (#4929)

Bug fixes

• Fixed a race condition where final task status could be overwritten by a late-arriving non-final status. (#4928)
• Conditional blocks no longer crash when evaluating empty or null parameters. (#4907)
• Browser sessions using FORCE_BROWSER_SESSION no longer time out prematurely. (#4903)
• SVG-based captcha text is now correctly extracted during DOM scraping. (#4880)
• MCP remote tool handlers now route to the correct API environment. (#4869)
• MCP remote auth token comparison no longer fails on encrypted tokens. (#4863)
• skyvern quickstart now handles local PostgreSQL without a pre-existing skyvern role. (#4878)
• WebSocket URLs through ngrok tunnels now rewrite correctly for live browser streaming. (#4927)

New features

• current_date Reserved Parameter — Workflows now have a built-in current_date parameter that resolves to today’s date automatically. (#4854)
• Reserved Parameters in Block Editor — The workflow block parameter picker now shows reserved system parameters, making them discoverable without memorizing names. (#4857)
• Natural Language Loop data_schema — Loop blocks driven by natural language extraction now support a data_schema field for consistent output structure. (#4851)
• Gemini 3.1 Pro and Inception Mercury-2 — Two new LLM engine options for your automations. (#4847)
• MCP Server on API Service — The MCP remote server is now mounted at /mcp on the existing API, eliminating the need for a separate process. (#4843)

Improvements

• Workflow save validation (HTTP 422) now returns clear, field-level error messages. (#4852)
• Browser launch errors now show actionable messages instead of raw exceptions. (#4844)
• Cloud LLM fallback overrides are now properly applied when selecting models. (#4839)

Bug fixes

• Conditional blocks no longer crash when all branches use Jinja templates. (#4849)
• Conditional blocks no longer misinterpret resolved Jinja template variables. (#4841)

New features

• Browser Profile Testing UI (Frontend) — The Browsers page now lets you launch a test run against a saved browser profile to confirm its login state is still valid before scheduling production workflows. (#4833)
• 2FA Verification Code UI and Notifications — When a workflow pauses for a 2FA code, the UI now shows a clear prompt and pushes a notification so you don’t miss it. (#4787)

Improvements

• 2FA Backend Cleanup — Refactored the 2FA pipeline to make code delivery more reliable across both the UI form and the API endpoint. (#4826)

New features

• Gated Admin Impersonation for MCP API Keys — Adds gated admin controls so support engineers can impersonate an organization scope when troubleshooting MCP-based automations, with audit-friendly access controls. (#4822)

Improvements

• Environment Variables Override Mounted API Keys — In Skyvern’s Streamlit-based deployment, env vars now take precedence over keys baked into the image, making local overrides simpler. (#4824)

New features

• Browser Profile Testing — Test browser profiles directly from the UI to verify that saved login sessions are still valid. Workflows can also run with a saved browser profile, enabling login-free automations that reuse authenticated sessions. (#4818)

Bug fixes

• When both a TOTP credential and a webhook are configured for 2FA, the credential is now correctly prioritized. (#4811)

New features

• Skyvern Skills Package — The new skyvern skill CLI command provides pre-built workflow templates and reference documentation for common automation patterns. (#4817)
• Clear Cached Scripts API — A new endpoint lets you clear cached automation scripts for a workflow when a target website changes and you want to force Skyvern to re-learn. (#4809)

New features

• Automatic 2FA Detection Without TOTP Credentials — Skyvern now detects when a website asks for a 2FA code even without pre-configured TOTP credentials. The system pauses and waits for you to provide the code via the UI or webhook. (#4786)
• Full CLI Parity with MCP — The CLI now supports browser session management, credential management, block operations, and enhanced workflow commands. Everything available through MCP is now accessible from the command line. (#4789, #4792, #4793)
• Claude Opus 4.6 CUA Support — Anthropic’s Claude Opus 4.6 is now available as a Computer Use Agent model for driving browser interactions. (#4780)

Improvements

• The workflow run timeline now properly displays loop iterations and conditional branches, making complex runs easier to debug. (#4782)

Bug fixes

• Fixed conditional blocks evaluating the wrong value after Jinja template rendering. (#4801)
• Fixed MFA resolution priority when both TOTP credentials and webhooks are configured. (#4800)

New features

• Anthropic Claude Opus 4.6 — Added Claude Opus 4.6 as an available LLM engine for web automation tasks. (#4777, #4778)

New features

• Unified Browser Task Block — The Navigation V1 and Task V2 blocks have been merged into a single Browser Task block. The block now exposes both engines as a dropdown, simplifying the workflow editor and reducing block-type confusion. (#4695)
• Block Discovery MCP Tools — Added MCP tools that let AI assistants enumerate available block types and their schemas, plus richer instructions for building workflows through MCP. (#4683)
• Edit Existing Credentials API — A new endpoint lets you update credential values without deleting and recreating them, preserving credential IDs across changes. (#4693)
• Filter Workflow Runs by Error Code and Search Key — The workflow runs list endpoint now supports error_code and search_key query parameters for narrowing down failed runs. (#4694)
• GPT-5 Mini in Model Selector — GPT-5 Mini is now selectable as an LLM engine. GPT-5 has been hidden in favor of more reliable variants. (#4686)

Improvements

• Block Failure Reasons Surfaced in UI — Every block type now shows its failure_reason in the run timeline, making debugging non-obvious failures faster. (#4672)
• Webhook Payload Truncation in Logs — Large webhook payloads are now truncated in log entries, keeping logs readable without losing context. (#4701)

Bug fixes

• Fixed the cursor jumping to the end while editing conditional expressions in the workflow editor. (#4682)
• Fixed dotted Jinja variables (e.g. {{ block_output.field }}) rendering incorrectly in the expression preview. (#4684)
• The API now returns 404 instead of 500 when a workflow is not found. (#4699)

New features

• Conditional Block Caching — The script cache now supports Conditional blocks, so workflows that branch based on page state can replay from cache instead of re-running the AI agent. (#4642)
• OpenTelemetry Backend — Skyvern can now export metrics, traces, and logs to any OTEL-compatible collector for centralized observability. (#4632)
• Workflow API Route Reorganization — Workflow API endpoints have been split into focused routers, making the OpenAPI surface easier to navigate. (#4628)

Improvements

• Workflow Copilot Auto-Opens for Empty Workflows — When you create a new workflow with no blocks yet, the Workflow Copilot opens by default to help you get started. (#4625)
• Run Workflow Button Moved to Top — The Run button now lives at the top of the workflow editor for easier access. (#4626)
• Skip LLM for Single-Match Location Autocomplete — Location autocomplete now skips the LLM call when only one option matches the user’s input, reducing cost and latency. (#4631)
• Improved Version History Icon — The workflow version history button now uses an icon that better represents document versions. (#4640)

Bug fixes

• Fixed ForLoop blocks failing to execute under script caching (regression from v1.0.12). (#4630)
• Fixed ActionDragPath subscript error in CUA drag-action parsing. (#4623)
• Fixed cron-scheduled tasks getting stuck with “task not found” errors. (#4627)
• Fixed parameter creation dialog not clearing default-value state when switching to credential type. (#4622)
• Fixed debugger loop-block visual overlap when the status row appeared. (#4641)

New features

• Task V2 Termination — Task V2 (the planner-agent-validator system) can now be terminated mid-run via API, matching the cancel behavior already available for Task V1. (#4589)
• ForLoop Block Caching — Cached scripts now support ForLoop blocks. Workflows that iterate over lists can replay from cache after the first successful run. (#4600)
• Conditional Block Caching — Initial support for caching workflows that include conditional branching. (#4603)
• run_with Parameter for Run Task — The run_task endpoint now accepts a run_with parameter so you can target Task V2 (or future variants) without changing the endpoint. (#4576)
• skyvern run dev Command — A new CLI command spins up a Superset workspace alongside the standard dev environment for local analytics work. (#4593)

Improvements

• All Branch Evaluations Shown in Conditional Block Timeline — The run timeline now shows every conditional branch’s evaluation result, not just the chosen one — useful when debugging “why did it pick that branch?” (#4577)
• Sidebar Expand/Collapse Icon — Replaced the pin icon with a chevron, matching common conventions. (#4586)
• PDF Parsing Token Cap — PDF parsing now caps token volume per request to avoid total failure on very large documents. (#4602)
• DATABASE_STRING Parameter on Quickstart — skyvern quickstart now accepts a DATABASE_STRING argument for explicit DB connection configuration. (#4580)

Bug fixes

• Fixed credentials not clearing from the Login block when deselected from the picker. (#4581)
• Fixed workflow run timeline missing recent steps until a manual refresh. (#4587)
• Fixed the workflow editor showing spurious “unsaved changes” when switching between conditional branches. (#4588)

New features

• Workflow Copilot Generally Available — The Workflow Copilot is now enabled for all users. Describe what you want in natural language and it generates a multi-block workflow you can edit. (#4533)
• Browser Sessions v2 Frontend — The new browser session management UI lands on the Browsers page, building on the v2 backend shipped in v1.0.10. (#4514)
• TOTP Endpoint OTP Type Filter — The TOTP listing endpoint now accepts an otp_type query parameter for filtering numeric codes vs. magic links. (#4529)
• Mintlify Documentation Setup — Skyvern’s documentation site moves to Mintlify, with redirects for the legacy Fern URLs. (#4516)

Improvements

• Async File I/O via aiofiles — Replaced blocking file I/O with aiofiles throughout the agent runtime, improving throughput on I/O-heavy workloads. (#4520)
• Cached Experimentation Provider TTL — BaseExperimentationProvider now caches feature-flag lookups with a per-item TTL, reducing pressure on the experimentation backend. (#4536, #4540)

Bug fixes

• Fixed ai_fallback not being read correctly from the database for cached workflow runs. (#4524)
• Fixed iframe-tree extraction crashing when the page raised a transient exception. (#4530)
• Fixed the workflow version comparison view ordering versions on the wrong sides. (#4531)
• Fixed the workflow history panel scroll behavior in debug mode. (#4532)
• Fixed TypeError when comparing offset-naive and offset-aware datetimes during workflow scheduling. (#4534)
• Fixed schema regeneration overwriting unchanged block field names. (#4535)
• Fixed uvicorn reload breaking on Windows due to async event loop policy. (#4538)

New features

• Browser Sessions v2 (Backend) — A new persistent browser session architecture with cleaner lifecycle management, better resource reuse, and improved reliability for long-running automations. The frontend follows in v1.0.11. (#4515)
• General Text CAPTCHA Solver — Skyvern can now solve a wider range of text-based CAPTCHAs without site-specific configuration. (#4517)
• TOTP Webhook Retries — TOTP webhook delivery now retries automatically on transient failures, reducing the rate of “no code received” errors during 2FA flows. (#4518)
• Block-Label Validation in run_blocks — The run_blocks endpoint now validates block labels up front, returning a clear error instead of failing partway through execution. (#4500)

Improvements

• Confirmation Dialog Shows Affected Blocks — Deleting a parameter or block now lists every block that references it, so you don’t accidentally break downstream logic. (#4519)
• Workflow Updates Return 409 on Race — Concurrent workflow updates now return HTTP 409 instead of silently overwriting each other. (#4510)
• Tiktoken Caching — Tokenizer instances are now cached, cutting per-step overhead on token counting. (#4521)

Bug fixes

• Fixed a Gemini 3 Flash cache-creation bug that caused false-positive fallback detection. (#4502)
• Fixed Bitwarden item-ID format validation rejecting valid IDs. (#4508)
• Fixed dangling block-output and parameter references after rename or deletion. (#4507)

New features

• Print PDF Block — A new workflow block that prints the current page to PDF and saves it to your run’s downloaded files, useful for archiving receipts, invoices, and reports. (#4452)
• HTTP File Downloads — The HTTP Request block can now download files directly via HTTP (not just GET JSON), enabling workflows that pull artifacts from authenticated APIs. (#4440)
• Workflow Copilot Streaming with Cancel — The Workflow Copilot now streams its responses with a cancel button, so you can stop generation mid-stream if you want to refine your prompt. (#4437, #4456)
• “Execute on Any Outcome” (Finally) Block Option — Blocks can now be configured to run regardless of whether prior blocks succeeded, enabling cleanup steps and “always send report” workflows. (#4443)
• Persistent Browser Session Uptime Billing — Billing v2 now meters persistent browser session uptime separately, with PostHog feature flags controlling rollout. (#4444)
• GPT-5 Mini Flex Engine — Added GPT-5 Mini Flex as an available LLM engine. (#4432)
• Clipboard Copy over HTTP — The browser now exposes clipboard copy functionality that works in HTTP-served browser sessions, not just HTTPS. (#4446)

Improvements

• HTTP Block Boolean and Integer Templates — The HTTP Request block now correctly passes boolean and integer template values without coercing them to strings. (#4435)
• Bitwarden and Azure Vault Tooltips — Added inline help tooltips explaining how to set up Bitwarden and Azure Key Vault credential sources. (#4447)
• Smaller S3 Objects Use STANDARD Tier — Storage cost optimization: small artifacts now use STANDARD instead of GLACIER_IR. (#4453)
• Reduced Max SVG Parsing — Lowered the SVG parsing cap to prevent runaway memory usage on pathological pages. (#4430)

New features

• Workflow Copilot v1 — The first version of the Workflow Copilot ships behind a feature flag. Describe a workflow in natural language and the copilot generates blocks, parameters, and connections you can edit on the canvas. (#4401, #4413, #4416, #4423)
• Per-Action Credit Billing (Billing v2) — A new billing model meters credits per browser action instead of per step, giving more predictable cost behavior on action-heavy workflows. (#4398)
• Azure Key Vault Secret Storage — Skyvern can now store and retrieve credentials from Azure Key Vault as a managed secret backend. (#4384)
• Conditional Inside Conditional — Conditional blocks can now be nested inside other conditional blocks for richer branching logic. (#4362)
• Reference Workflow Output in Sub-Blocks — Blocks can now reference upstream workflow_output variables, simplifying multi-stage workflows. (#4414)
• Browser Type Choice for Persistent Browser Sessions — Persistent sessions now let you choose the browser type (Chromium, etc.) at create time. (#4406)
• Browser Extension Support for PBS — Persistent browser sessions can launch with a specified browser extension. (#4364)
• Step Count in Webhooks and Run Payloads — Webhook payloads and get_run responses now include step_count for cost reconciliation. (#4410)
• PH (Philippines) Proxy Location — Added Philippines as a residential proxy option. (#4395)
• TextPromptBlock Artifact Storage — TextPromptBlock now saves the LLM prompt and response as run artifacts for debugging. (#4361)
• Action–Screenshot Linking — Actions in the run timeline now link directly to the screenshot captured at the time of execution, so you can see what the agent saw when it made each decision. (#4403, #4404)

Improvements

• Clearer Errors for Missing Workflow Parameters — Missing required parameters now return a clear HTTP 400 instead of HTTP 500. (#4380)
• Empty-String Defaults for String Parameters — Workflow string parameters without explicit defaults now initialize to "" instead of null, matching common Jinja-template expectations. (#4382)
• HAR Files Use Zstd Compression — HAR archives now ship as .har.zst, cutting storage size significantly with no fidelity loss. (#4420)
• Hydrated Workflow Title in Run Responses — workflow_run responses now include the workflow title for cleaner logging without an extra API call. (#4363)
• Cached Workflow-Script Lookup — get_workflow_script_by_cache_key_value is now cached and post-action execution can re-enable code generation. (#4385)
• Updated Datacenter Proxy Provider — Datacenter proxies now route through a new provider with broader IP coverage. (#4386)
• Improved Conditional NL Branch Tooltip — Updated tooltip and placeholder text on conditional blocks to better describe how natural-language branches are evaluated. (#4368, #4369)
• OLLAMA Vision Support Toggle — Added OLLAMA_SUPPORTS_VISION env var to enable vision-capable Ollama models. (#4351)

Bug fixes

• Fixed older credentials becoming inaccessible after an internal schema change. (#4358)
• Fixed scrolling actions misfiring on pages with overflow-hidden ancestors. (#4389)
• Fixed close_browser_on_completion not being honored during workflow run cleanup. (#4381)
• Fixed run-ID generation collisions under high concurrency. (#4383)
• Fixed empty browser-extension config crashing PBS launch. (#4365)
• Fixed wrong screenshots being attached to certain actions. (#4400)
• Fixed nested DB connection management in workflow execution. (#4402, #4411)
• Fixed local variable not defined errors in cached-action execution. (#4376)
• Fixed artifacts endpoint incorrectly including api/ in the path. (#4417)
• Fixed S3 storage class selection for small objects (_get_storage_class_for_org). (#4424)
• Fixed cached actions executing without a brief settling delay. (#4418)