AI Coder Releases

​

🎯 Highlights

• Major HIPAA agent speed and reliability upgrade with parallel sub-agents
• Deep security scanning: PHI leakage detection for payment providers, browser storage, audit logs, and more

​

✨ New Features

• HIPAA Agent: Parallel Scanning: Scans now run with sub-agents for significantly faster results
• HIPAA Agent: PHI Leakage Detection: New checks catch PHI leaking to payment providers (Stripe), session tokens stored in browser storage, clinical file links that never expire, and PHI in audit log fields or CSV exports
• HIPAA Agent: Endpoint Security: Detects unprotected seed/reset/debug endpoints, unverified webhooks, and unprotected admin/cron routes
• HIPAA Agent: Build Health Context: Scan results now show build and config health status as context banners

​

🔧 Improvements

• HIPAA Agent: 2026 Rule Calibration: Pre-enforcement rules now treated as advisory suggestions rather than hard requirements
• HIPAA Agent: Smarter Access Scoping: Distinguishes intentionally broad provider access from unscoped access to reduce false positives
• HIPAA Agent: Security Headers: Advisory flags for missing security headers
• HIPAA Agent: Form Abuse Protection: Flags public forms that create records or send email without rate limiting or CAPTCHA
• HIPAA Agent: Appointment Email Calibration: Stopped flagging compliant appointment reminder emails (aligned with HHS guidance)
• HIPAA Agent: Server-Side Auditing: PHI read auditing now required to happen server-side, not from the browser

​

🐛 Bug Fixes

• Fixed HIPAA agent not clearing resolved issues from scan results
• Fixed compliance score not improving after multiple fix-and-rescan cycles

​

🎯 Highlights

• Faster, less verbose agent responses with better post-action guidance

​

✨ New Features

• Design Review Flow: Design changes now require explicit approval before the dev agent implements them

​

🔧 Improvements

• Less Verbose Agents: Reduced agent verbosity for cleaner, more focused responses
• Post-Design CTA: After the design agent finishes, users now get a clear call-to-action to continue
• HIPAA Agent Fix Detection: Scanner now properly picks up fixes applied in subsequent scans

​

🐛 Bug Fixes

• Fixed 2-3 second delay between hitting enter and the message being sent
• Fixed external git repo imports (from Lovable, etc.) being incorrectly accepted
• Fixed agent falsely claiming “Changes uncommitted on main”
• Fixed git push failing in workspaces due to invalid credentials
• Fixed “error in project” message showing when no project exists yet
• Fixed subagent work getting cut off and restarted, prolonging build slices
• Fixed git HEAD reset destroying verified commits mid-session

​

🎯 Highlights

• Comprehensive HIPAA scanning overhaul with 10+ new security check categories

​

✨ New Features

• HIPAA Agent: Expanded Coverage: New scan categories including RBAC detection, audit log detection, data handling and display, file uploads and documents, data encryption (at rest and in transit), API security, infrastructure and environment, PHI exposure in UI/URLs, and third-party services and BAAs

​

🐛 Bug Fixes

• Fixed AI responses appearing duplicated in chat
• Fixed HIPAA compliance agent failing with module errors in production
• Fixed AI coder leaking internal chain-of-thought into user-facing chat
• Fixed false 524 deployment errors during Convex deploy
• Fixed callback hook blocking implementation progress when server showed no pending reviews

​

🎯 Highlights

• New CometChat component for in-app messaging

​

✨ New Features

• Maestro: Audit Log Implementation: Maestro can now plan and implement audit logging for PHI/ePHI surfaces
• Maestro: CometChat Skill: New integration skill for adding CometChat in-app messaging

​

🔧 Improvements

• HIPAA Agent UI: Updated button styling to match primary CTAs across the app

​

🐛 Bug Fixes

• Fixed account seeding not working properly - agent now shows all created accounts
• Fixed planning questions not appearing, leaving the app stuck without asking initial questions
• Fixed repetitive Vite build errors during development
• Fixed roadmap not being shown by the agent after planning
• Fixed questionnaire questions disappearing before users could answer, showing “Waiting on answers” prematurely

​

🎯 Highlights

• New Maestro skill system: 8 healthcare-specific skills (Intake, Scheduling, Admin UX, Provider Profile, Provider Search, Organization Topology, MFA Implementation, Twilio SMS)

​

✨ New Features

• Maestro Skills: Intake, Scheduling, Admin UX, Provider Profile, Provider Search, Organization Topology, MFA Implementation, and Twilio SMS skills
• HIPAA Agent MFA Scan: New MFA compliance scanning capability

​

🔧 Improvements

• Maestro: First build slice + skills: Improved how first build works with the skill system
• Maestro: Design agent image guidance: Design agent now clearly calls out when users should upload images to steer design
• Maestro: No image upload folder instructions: Stopped asking users to upload images to specific folders
• Maestro: Mobile apps scope guardrail: Mobile apps flagged as out of scope in MVP, require custom plan
• Maestro: Skill hot-reload: Kill instances so skills reload when updated to fix bugs
• Rate limit increase: Increased rate limit with E2B

​

🐛 Bug Fixes

• Rate limit deploy error: Fixed rate limit error when deploying - must be opened in a deployed preview, won’t work in the inline preview
• CSS @import ordering: Fixed ECONNREFUSED during design phase caused by CSS @import ordering
• HIPAA agent failures: Fixed HIPAA agent run failing intermittently in production
• Duplicate planning messages: Fixed Maestro showing duplicate planning message/duplicate questions
• “Message and product name required” error: Fixed AI coder error message
• Stripe Integration Preview: Fixed preview never loading for Stripe integrations
• Bizarre “request failed” error: Fixed mysterious request failed error
• Table UI in chat: Fixed weird table UI rendering in AI chat responses
• Roadmap/Design brief visibility: Fixed roadmap and design brief sometimes not showing up

​

🎯 Highlights

• Free trial credits increased to 80 (up from 50)
• Improved preview reliability after builds

​

✨ New Features

• Clone Projects: “Duplicate” renamed to “Clone” with a confirmation dialog to prevent accidental copies

​

🔧 Improvements

• More Free Trial Credits: Free trial allowance increased from 50 to 80 credits, giving new users more room to explore
• Better Seed Account Display: After seeding, the agent now shows all created accounts instead of just a single password
• Stripe Webhook Guidance: Stripe integration now tells you exactly which events the webhook needs to listen for
• Smarter Image Sourcing: The agent now correctly uses images from Unsplash/Pexels in designs

​

🐛 Bug Fixes

• Fixed roadmap not appearing after planning completes successfully
• Fixed preview not auto-loading after first build completes
• Fixed error display overlapping mobile preview - errors now show above the preview window

​

🎯 Highlights

• Smarter planning experience with improved question flow
• Significant UI polish across chat, preview, and developer console
• Better error handling so you spend less time guessing what went wrong

​

✨ New Features

• Mailgun Integration: New one-shot integration skill for Mailgun email delivery
• Page Refresh Recovery: If you refresh mid-build, Specode now shows an indicator that the agent is still working

​

🔧 Improvements

• Better Branding & Design: Improved color picking and branding suggestions during the design phase, with guidance to upload images for best results
• Improved First Slice Scope: First builds now include a more complete starting point beyond just a landing page
• Smarter Log Handling: The agent checks logs directly instead of telling you to check them yourself
• Focused First Slice: Integrations are no longer added prematurely in the first build slice
• Reduced Deferred Scope Hallucination: The planner is more grounded in what it defers to later phases
• Readability on Blurred Preview: Lighter overlay text color for better readability in both dark and light mode

​

🐛 Bug Fixes

• Fixed long AI messages breaking the chat layout
• Replaced text upload indicator with a clean circular progress indicator
• Fixed chat not fully scrolling to the bottom when streaming messages
• Fixed patient input getting cut off when the sidebar is narrow
• Fixed dev console UI layout issues
• Fixed card width in chat and repositioned the action menu
• Fixed sandbox error briefly flashing before the blur overlay loads
• Various UI refinements for preview panel and header
• Fixed design agent response disappearing mid-conversation
• Fixed spammy infinite-loop message sequences from the agent
• Fixed planning questions not appearing after the planning agent completes
• Fixed double streaming during agent handover
• Fixed first slice confirmation message disappearing from chat after being written
• Fixed agent re-asking for design approval after it was already given
• Fixed preview showing blank white screen instead of auto-refreshing after first slice
• Fixed Stripe integration showing contradicting API key data and missing webhook URL
• Fixed auth seeding failures
• Fixed Cloudflare error preventing the agent from starting when using tools

​

🎯 Highlights

• Introducing Maestro: a new conversational workflow that takes you from a simple prompt to a working app in minutes
• Redesigned planning, design, and build phases for a smoother experience

​

✨ New Features

• Conversational Planning Flow: Describe your app idea in a simple prompt and Maestro generates a full PRD through a guided conversation
• Design Phase: Review and approve a design brief before building begins
• First-Shot Build: Get a working 3-5 screen app in ~10 minutes from your approved plan
• Conversational Iteration: Refine your app through natural back-and-forth with the AI coder
• Multi-Image Upload: Upload multiple reference images at once during planning
• Example Login Credentials: Auth-enabled apps now come with seed logins so you can test different roles immediately
• HIPAA Compliance Responses: The AI agent can now answer HIPAA compliance questions directly
• Legacy Project Support: Existing projects work seamlessly with the new workflow

​

🔧 Improvements

• Default 50/50 split between chat sidebar and app preview
• App preview opens immediately - no more dashboard loading spinner
• New preview controls: hide, launch in new tab, and refresh
• Planning conversations are higher quality and more focused
• Design agent creates a focused MVP scope

​

🐛 Bug Fixes

• Fixed AI agent losing chat context between planning, design, and build steps
• Fixed AI not responding in chat after initial build (work continued in background but chat appeared stuck)
• Fixed code preview showing placeholder text during the planning phase
• Fixed delay/no feedback when sending longer messages
• Fixed strange sandbox messages about “agent process starting” appearing in chat
• Fixed agent incorrectly asking users to run Convex commands for API keys
• Added automatic retry for GitHub errors

• Run the agent on demand from the Compliance Center inside the AI Coder whenever you want to check your app’s compliance
• The agent uses a multi-agent system: two independent AI agents scan your code in parallel, then a third merges and verifies the results to reduce false positives
• Issues are categorized by severity (Critical, High, Medium, Low) so you know what to fix first
• Once flagged, you can copy the issue into the AI Coder chat and ask it to fix it for you
• Re-run the agent after fixes and it will automatically mark resolved issues as complete

• Each full scan takes roughly 3-4 minutes and costs approximately 1 credit
• The agent includes automatic retry if a scan fails mid-run
• All scan history is saved so you can track your compliance progress over time

​

🎯 Highlights

• Pen testing is now available as a purchasable service directly within Specode
• Pro plan upgraded with 2,200 credits (nearly 3x), hands-on support, and expanded integrations

​

✨ New Features

• Pen Testing Purchase Flow: Purchase an optional one-time pen testing service ($3,000) directly through Specode. Includes a 90-day window for re-tests.
• AI Coder Observability: Added monitoring and logging for AI coder sessions so we can diagnose and resolve errors faster when sessions reset or fail to respond.
• Expanded Pro Plan: Pro now includes 2,200 credits, hands-on support with prompt guidance and unblocking sessions.

​

🔧 Improvements

• Design Quality Upgrade: Added a design skill to the AI coder so generated apps produce cleaner, more polished UIs out of the box.
• Faster Subscription Activation: Reduced wait time when activating or renewing a subscription.
• Deploy Errors Surfaced in UI: Preview deployment errors are now shown directly in the app instead of requiring users to check the browser console.
• Forgot Password Works Out of the Box: Password reset now uses default email credentials so it works immediately without needing to set up a third-party email service first.
• Submit Login with Enter Key: Login form now submits when you press Enter.

​

🐛 Bug Fixes

• Fixed an issue where new customers saw a loading screen after submitting payment with no confirmation (payment still processed in background).
• Fixed linter errors not being surfaced before deployment, which caused users to hit errors only after attempting to deploy.

​

🎯 Highlights

• Significantly faster project loading and setup times
• Fixed critical authentication and connectivity issues

​

✨ New Features

• Dynamic Cross-Domain Authentication: Signing into your deployed Specode app preview now works seamlessly across all environments, resolving redirect errors that previously occurred on Vercel previews

​

🔧 Improvements

• Faster Project Loading: Projects now load noticeably faster thanks to parallel processing, optimized dependencies, and smarter caching - saving up to a minute in some cases
• Smoother Startup Experience: Eliminated the blank white screen that appeared when opening a project
• Security Update: Upgraded our underlying framework to address a critical security vulnerability

​

🐛 Bug Fixes

• Fixed an issue where environment variables set in your project weren’t being recognized
• Fixed the HIPAA compliance banner incorrectly appearing in the AI coding workspace instead of only in app previews
• Fixed the “Reload this page” popup that would interrupt you while the AI was working on your app
• Fixed a “Refused to connect” error that occurred when logging out and back in on certain browsers

​

🎯 Highlights

• Complete platform rebuild on serverless functions for backend for dramatically improved performance
• More powerful Smart AI Agent
• New pay-as-you-go credit system for flexible usage
• Faster, semi-automated HIPAA-compliant deployment pipeline

​

✨ New Features

• New Tech Stack: Migrated to serverless backend, replacing Supabase + Render infrastructure with approximately $900 USD/month savings
• Advanced AI Model Addition: AI Coder now powered by the most advanced AI models
• Pay-As-You-Go Credits: Flexible credit purchasing system and smart consumption logic
• Dev Console: New console providing real-time activity transparency and debugging capabilities
• Interactive Navigation Bar: Redesigned navigation for faster project workflows
• Preview Modal Redesign: Updated preview button and modal with HIPAA warning banner for compliance awareness
• Publish Flow: New deployment flow - Submit your deployment to the support queue
• Logo & Favicon Upload: Upload custom branding assets directly in the editor
• Video Tutorial Playlist: YouTube playlist for onboarding
• Custom Plan Credits: Enterprise support for custom monthly credit allocations

​

🔧 Improvements

• Massive Performance Update: Significantly snappier UX, faster page generation, quicker code changes, and fully stable previews
• Web Container Stability: Resolved disconnection and stability issues
• AI Response Formatting: Improved thinking sequence display
• Blank Template: Introduced a Blank Template at initial project creation
• Auth Template: Blank template with authentication built in

​

🐛 Bug Fixes

• Dashboard 404 Error: Fixed refresh (cmd+r) causing 404 on Vercel-hosted dashboard
• Port 5174 Error: Resolved “Unable to connect to port 5174” after Vercel previews
• Video Player UI: Fixed styling and call-to-action to match v1
• Project Creation Spinner: Added loading indicator to prevent “stuck” appearance
• Free Tier Error Message: Updated messaging for users attempting to create more projects
• Legacy Modal (v1 Projects): Fixed display for existing users loading v1 projects
• Legacy Modal (First v2 Project): Fixed display for existing users creating first v2 project

​

🔄 Legacy Support

• v1 project routing and support maintained for paid users backwards compatibility

​

🎯 Highlights

• Interactive example prompts for improved onboarding experience
• New admin section for platform management
• Enhanced navigation with help button

​

✨ New Features

• Prompt Example Suggestions: Interactive example prompt feature that cycles through real-world healthcare use cases. Users can click “See example” to explore different prompts that automatically type out the prompt.
• Admin Section: New administrative interface for specode.ai platform management
• Help Button: Added help button to top navigation for easier access to support resources

​

🔧 Improvements

• Prompt Text Area Styling: Updated text color in prompt text area to use primary text color for better readability

​

🐛 Bug Fixes

• AI Coder Chat Truncation: Fixed issue where AI Coder was cutting off prompt chats prematurely

​

🎯 Highlights

• Initial production release of Specode AI Coder and Foundation platform
• Comprehensive EMR modal components for allergies, conditions, and medications
• Enhanced AI Coder chat interface with improved output formatting
• Production-ready deployment

​

✨ New Features

• EMR Allergy Management: Add/Edit modal component for patient allergies
• EMR Condition Management: Add/Edit modal component for patient conditions
• EMR Medication Management: Add/Edit modal component for patient medications

​

🔧 Improvements

• Chat Output Formatting: Improved AI Coder response display with better structure and reduced delays between message segments
• Environment Setup: Environments configured on Supabase

​

🐛 Bug Fixes

• Fixed AI Coder project loading failures
• Resolved connection interrupted errors
• Fixed intermittent network errors that required page refresh
• General UI consistency improvements and cleanup