Sasha Studio Release Notes: v1.0.1068 to v1.0.1121
Release Period: January 2026 - February 2026
Version Range: 1.0.1068 → 1.0.1121
Total Commits: 140
Actual Code Changes: 344 files changed, 13,144 insertions, 1,808 deletions
TL;DR - Business Summary
What's New in Plain English:
This release delivers 5 major capabilities that make Sasha smarter, more organized, and more interactive:
Interactive Prompt Cards - When Claude asks you a question during a conversation, you now see clickable option cards instead of plain text. Pick your answer with a tap instead of typing it out.
Knowledge Tags & Filtering - Tag your knowledge documents with categories and filter them instantly. Find what you need without scrolling through long lists.
Resizable & Fullscreen Previews - Drag-resize the sidebar and doc preview panel to your preferred width. Go fullscreen on document previews for deep reading.
PostHog Analytics - Product analytics integration gives you visibility into how your team uses Sasha, enabling data-driven decisions about training and adoption.
SEO & Public Pages - New public about page with sitemap and robots.txt for search engine visibility, helping prospective users discover your Sasha instance.
Business Value:
- Productivity: Interactive prompts reduce conversation friction — answer Claude's questions with one click
- Organization: Tags and sorting modes make large knowledge bases navigable
- Adoption: Analytics show which features your team actually uses
- Discoverability: SEO improvements help new users find your Sasha instance
Executive Summary
This release transforms how users interact with Claude's questions and how they organize knowledge. The headline feature is Interactive Prompt Cards — a complete redesign of how Claude's AskUserQuestion and ExitPlanMode tools appear in the chat. Instead of raw text output, users now see styled, clickable cards with clear options. This required deep changes to the streaming pipeline, WebSocket handlers, and message reconciliation system.
The Knowledge Management system received substantial upgrades: frontmatter-based tags with inline editing, multiple sort modes (alphabetical, date, type), file-type-specific icons, and smart file display logic that shows HTML previews instead of raw markdown when available. A deploy-time backfill script adds created dates to all existing documents.
On the infrastructure side, PostHog analytics integration provides product usage insights, a new memory monitor utility prevents OOM situations, and 78 Dependabot security alerts were resolved through dependency overrides. The system prompt injection system was formalized with a registry and structured logging, making it easier to audit and maintain the invisible prompts that shape Claude's behavior.
Major Features & Improvements
Interactive Prompt Cards
- AskUserQuestion Rendering - Claude's questions now display as styled interactive cards with clickable option buttons
- ExitPlanMode Card - Plan review prompts render with expandable plan content and blue/gray design system
- Streaming Integration - Interactive prompts emit via
claude-interactive-promptevents, wired into active WebSocket handlers - Deduplication - Prompt cards are deduplicated after history reconciliation to prevent duplicates on page reload
- Streaming Order Fix - Interactive prompts no longer appear before or replace streaming text content
- Suppressed Fallbacks - Claude's text-based fallback output and error
tool_resultsare suppressed when interactive cards are active
Knowledge Tags & Organization
- Frontmatter Tags - Add tags to knowledge documents via YAML frontmatter (
tags: [sales, onboarding]) - Inline Tag Editing - Right-click any document in the sidebar to edit its tags
- Tag Filtering - Filter the Knowledge sidebar by tags with a collapsible tag selector (+N popover for overflow)
- File Manager Tags - Tag filter system also available in the File Manager view
- Sort Modes - Toggle between alphabetical, date-modified, and file-type sorting in the Knowledge sidebar
- File-Type Icons - Documents display type-specific icons (PDF, CSV, HTML, MD, etc.) instead of generic file icons
- Smart File Display - Sidebar shows HTML preview versions instead of raw markdown when available; shows .md files in file manager even when HTML counterpart exists
- Created Dates - All markdown file creation paths now inject
createdfrontmatter; deploy-time backfill script adds dates to existing documents
UI/UX Improvements
- Resizable Sidebar - Drag handle lets you resize the sidebar to your preferred width
- Resizable Doc Preview - Document preview panel is independently resizable
- Fullscreen Preview - Toggle document previews to fullscreen for focused reading
- Collapsible Bottom Action Bar - Action bar collapses when not needed
- Truncatable User Messages - Long user messages can be collapsed with configurable settings
- Plan Review Cards - Open by default with unified blue styling and expand/collapse
- Trailing Question Highlighting - Questions at the end of assistant messages are highlighted with a blue block
- Show Work in Progress Toggle - Quick Settings toggle to show/hide in-progress indicators
- Document Search in History - Search palette now includes document search results
Analytics & SEO
- PostHog Integration - Product analytics for tracking feature usage and user behavior
- Response Types Chart Improvements - Value labels and y-axis labels added to analytics charts
- Historical Session Data - Host activity report now includes historical session data
- Public About Page - SEO-optimized public page for search engine discovery
- Sitemap & Robots.txt - Standard SEO files for search engine crawling
- Project Limit Setting - Historian project limit now configurable from the UI
System Architecture
- System Prompt Injection Registry - Formalized registry with structured
📋 [SYSTEM-PROMPT]logging for all injections - AskUserQuestion Injection - New system prompt injection that enables Claude to render interactive question cards
- Isolated Temp Directories - Each Claude CLI process gets its own temp directory to prevent cross-session file conflicts
- GitHub Token Seeding -
SASHA_GITHUB_TOKENautomatically seeded into secretstore on boot - Memory Monitor Utility - New utility for tracking and alerting on memory consumption
- SIGTERM/143 Diagnostics - Comprehensive error diagnostics when Claude processes are terminated
Stability & Reliability
Streaming & Message Pipeline
- Object Serialization Fix - Prevented
[object Object]loss in the message pipeline by usingJSON.stringify - Chunk Deduplication - Consecutive identical chunks are deduplicated in the streaming buffer layer
- Final Output Fix - Resolved issue where final output was missing until page refresh
- Tool Result Indicators - Gray indicator shown for tools with missing results in historic sessions
- Thinking Indicator - Respects
showThinkingsetting for in-progress indicators
Knowledge System
- DOCS_ROOT Search Path - Added to conversation docs search paths for reliable file discovery
- CSV/HTML/TXT Support - Sidebar now shows these file types alongside markdown
- HTML Preview Whitelist - HTML added to raw previewable extensions
- Simplified File Discovery - Streamlined to include project root
Security
- 78 Dependabot Alerts Resolved - Dependency overrides for known vulnerabilities
- Health Endpoint Auth - Authentication added to health check API calls
- MCP Health Check Consolidation - Startup MCP checks consolidated to prevent OOM from concurrent checks
- JWT Error Logging - Improved logging for secret rotation scenarios
- Per-Client Health Secrets - Host activity reports use per-client health check secrets
Process Management
- Claude Historian Sort Fix - Projects sorted by mtime to find most recent sessions
- Absolute Paths for Uploads - @ references on uploaded files now use absolute paths
- Multi-line JSON Parsing - Stats handle multi-line JSON in events.jsonl
Developer Experience & Docs
Documentation Updates
- System Prompt Injection Registry - Complete documentation of all 6 active injections with mandatory process for adding new ones
- Interactive Prompt Cards - Architecture and streaming integration documentation
- Interactive Prompt Redesign - Design spec for the blue/gray card system
- Knowledge Tags - Feature documentation for the tagging system
- Knowledge Frontmatter Dates - Documentation for the created-date backfill system
- Knowledge Sidebar Sort Modes - Documentation for sorting feature
- Knowledge File System - Updated file discovery documentation
- Message Pipeline - Documentation for the message serialization pipeline
- Claude CLI Tool Result Handling - Technical documentation for tool result suppression
- PostHog Integration - Analytics setup and configuration guide
- Host Activity Reporting - Feature documentation
- Streaming Order Decision - ADR for interactive prompt streaming order (issue #438)
New Features for Developers
- Start New Project - "Start new project" option in folder context menu
- Navigation to Chat - Top menu bar Chat button now navigates properly
Upgrade Notes
Interactive Prompt Cards
- Automatic: All
AskUserQuestionandExitPlanModeresponses from Claude now render as interactive cards - No Configuration Needed: Works with existing conversations
- Fallback: If card rendering fails, plain text output is still available
Knowledge Tags
- Backfill Script: Run
node scripts/add-frontmatter-dates.jsto addcreateddates to existing documents - Deploy-Time: Backfill runs automatically during Docker deployment
- Tags Format: Add
tags: [tag1, tag2]to document frontmatter
PostHog Analytics
- Configuration: Set PostHog API key as build argument for Docker (
POSTHOG_KEY) - Privacy: Analytics are opt-in and respect user preferences
- No User Action Required: Analytics are disabled by default without a key
Breaking Changes
- None: All changes are backward compatible
Changelog Summary (since v1.0.1068)
Features
- Interactive prompt cards for AskUserQuestion and ExitPlanMode
- Knowledge frontmatter tags with inline editing and filtering
- File-type-specific icons in Knowledge sidebar
- Toggleable sort modes (alpha, date, type) in Knowledge sidebar
- Resizable sidebar and doc preview panel with drag handles
- Fullscreen toggle for document preview
- Collapsible bottom action bar
- Truncatable user messages with configurable limits
- Document search in history search palette
- Plan review card redesign with blue/gray system
- Trailing question highlighting in assistant messages
- "Show work in progress" toggle in Quick Settings
- PostHog analytics integration
- Public about page with sitemap and robots.txt
- System prompt injection registry with structured logging
- AskUserQuestion system prompt injection
- Isolated temp directories per Claude CLI process
- GitHub token seeding into secretstore on boot
- SIGTERM/143 diagnostic messages
- Configurable project limit for historian
- PDF download from preview panel
- Historical session data in host activity reports
- Value labels and y-axis on Response Types chart
- "Start new project" in folder menu
- File rename in Knowledge sidebar context menu
- Support renaming MD-generated HTML files via source markdown
Fixes
- 78 Dependabot security alerts resolved via dependency overrides
[object Object]serialization loss in message pipeline- Streaming chunk deduplication in buffer layer
- Final output missing until page refresh
- Interactive prompt deduplication after history reconciliation
- Interactive prompt streaming order (appears after text, not before)
- Claude text fallback suppression with interactive prompts
- Tool result error suppression from CLI stream
- MCP health check consolidation to prevent OOM
- Health endpoint authentication
- Per-client health check secrets for host activity
- Knowledge sidebar file-type icons applied correctly
- MD files hidden in sidebar when HTML version exists
- CSV, HTML, TXT files shown in Knowledge sidebar
- DOCS_ROOT added to conversation docs search paths
- HTML added to previewable extensions whitelist
- Claude historian projects sorted by mtime
- Absolute paths for @ references on uploaded files
- Multi-line JSON parsing in events.jsonl
- Sidebar resize handle hidden during fullscreen
- Navigation added to Chat menu button
- Knowledge articles refresh on section expand
- Streaming message repetition prevention
- Gray indicator for tools with missing results in history
- showThinking setting respected for in-progress indicator
Looking Ahead
- Smart Tag Suggestions: AI-powered tag recommendations based on document content
- Interactive Prompt History: Browse and revisit past prompt card interactions
- Knowledge Dashboard: Visual overview of document coverage, tags, and freshness
- Team Analytics: Usage patterns by team and role via PostHog
- Document Relationships: Automatic linking between related knowledge articles
Jargon Buster - Technical Terms Explained
Interactive Prompt Cards
- Clickable UI cards that appear when Claude asks you a question during chat
- Like a multiple-choice form embedded in the conversation
- Replaces plain text questions with a visual, click-to-answer interface
Frontmatter
- Metadata at the top of a markdown file between
---markers - Like the label on a folder — contains tags, dates, and other organizational info
- Used by Sasha to add tags and creation dates to knowledge documents
System Prompt Injection
- Invisible instructions sent to Claude alongside your conversation
- Like whispering context to an assistant before they talk to a customer
- Controls Claude's behavior (e.g., persona, guides, plan mode)
WebSocket
- A persistent connection between your browser and the server for real-time updates
- Like keeping a phone line open instead of hanging up after each sentence
- Enables streaming responses and interactive prompt card delivery
PostHog
- Open-source product analytics platform
- Like Google Analytics but for understanding feature usage within your app
- Tracks which Sasha features your team uses most
Dependabot
- GitHub's automated security scanner that flags vulnerable dependencies
- Like a security guard checking your building's suppliers
- 78 alerts were resolved by updating or overriding affected packages
OOM (Out of Memory)
- When a process tries to use more memory than available
- MCP health checks were consolidated to prevent too many running at once
- Memory monitor utility now provides early warnings
SIGTERM/Signal 143
- An operating system signal telling a process to shut down
- Like a manager tapping someone on the shoulder to wrap up
- New diagnostics explain why a Claude process was terminated
Frontmatter Backfill
- Adding metadata to existing documents that didn't have it
- Like going through old filing cabinets and stamping dates on unlabeled folders
- Runs automatically during deployment to ensure all docs have creation dates
Thanks for upgrading. This release makes Sasha more interactive with clickable prompt cards, more organized with knowledge tags and sorting, and more observable with analytics — all while resolving 78 security alerts and hardening the streaming pipeline.