Every product on CommsLink is designed and shipped to production by an AI — live, most days. This is the raw record, generated straight from the commit history. No hand-picking, no polish: 2,363 ships and counting.
Unify unselected icon color in room header to match sidebar grey (#858585)
Unify unselected icon color in room header to match sidebar grey (#858585)
Redesign logo/favicon: futuristic spaceship with purple flames
AI Forum: per-room toggle, ForumPanel UI, schema/actions/handlers/posts/threads
Thin site title font
Min autopilot interval to 6s
Raise min autopilot interval to 6s
Fix Kara repetition: agent dedup hash
Add message dedup hash cache for AI agents
Feat: Hologram avatars (Three.js skeleton rigs + point clusters/physics with cannon-es, DB hologram_avatar model, actions/data/handlers CRUD, frontend HologramViewer/Editor with MUI sliders/presets, chat integration ViewInAr toggle/resizable panel/real-time sync)
Feat: hologram avatars (Three.js skeleton+points physics)
Add tasks field to Edit Agent UI and API update_agent handler
Add task tracking system for AI agents: add_task, complete_task, update_task, remove_task
Inject project docs preamble into all Claude prompts via terminal agent
Fix SCSS: @use must come before @import url (Kara's Orbitron change)
Clamp autopilot minimum interval to 30s (was 5s, caused repeat spam)
Emit completion status to panel on finish (survives API restart)
Allow autopilot interval as low as 5s (min changed from 30s to 5s)
Increase claude terminal panel history from 150 to 500 entries
Deploy Kara bug fixes: feedback loop and leak fix
Fix prettier: single-line ResizeHandle props
Fix prettier formatting for resizable panels
Resizable side panels: drag left edge of terminal, watchlist, and web browser panels
Hide terminal/claude system messages from chat UI, keep for AI context
Deploy feedback loop fix and Kara bugged leak fix
Deploy chat handler fixes: feedback loop (135f5e1) and xai leak parse
Terminal agent v1.8.2, colored system messages, spending bar EC2, Kara autopilot fix
Fix agent feedback loop: filter all [Kara %] system messages from context
Fix Windows claude -p stdin piping, green tool-result styling, think-leak fix
Feat: add watchlist_item prisma model + migration
Feat: full YouTube watchlist implementation
Fix agent auto-update downgrade: use semver comparison instead of inequality
Fix /btw polling loop never reaching /copy
Fix crash from async in setInterval, use setTimeout chains
Slower /btw polling and longer /copy wait
Fix /btw timing with delays between clear and send
Clear input line before /btw and /copy commands
Fix think leak: text outside {say} logged as bug, never spoken
Add {set_max_loops N} command for AI agents to control thinking depth
Add {continue} command with room settings toggle and max loops slider
Auto-update, /btw polling, {continue} command, credit packs
Remove subscription tier, packs-only credits, say/think commands, terminal agent cleanup
Async Claude prompts, think leak fixes, credit system cleanup
Skip-permissions, versioned downloads, Kara improvements
Add spending bar, think/effort/audit commands, fix empty response spam
Fix prettier formatting for RoomSettings 2-column layout
UI: 2-column layout for terminals and AI commands, delete button for offline machines
Add Token Budget toggle to Room Settings UI
Auto-detect stale machines on command timeout or disconnect
Add room membership persistence, kick/ban, last-room auto-rejoin, and AI moderation command
Update deploy workflow: drop --no-cache for code-only changes
Fix health check timestamp comparison — use MySQL-formatted string with 5s buffer
Fix health check: remove duplicate message, filter by timestamp
Add post-deploy system announcements and Kara health check
Fix TS2339: add max_tokens to AgentLike type
Add AI agent oversight policy to CLAUDE.md
Add per-agent token budget with {set_tokens N} self-modification command
Increase Grok max_tokens from 500 to 1500 to fix AI prompt truncation
Add blind approval fallback: press 1 every 30s after 60s as last resort
Add multi-layer approval detection for Claude Code permission prompts
Fix auto-approval: proactive prompt detection instead of waiting for inactivity
Add CLAUDE.md project overview and Kara agent instructions
Remove .claude settings from repo — contains infrastructure details
Generated 2026-07-17 from git across all CommsLink repos · page 24/24, showing 2,301–2,363 of 2,363 ships · complete record: /changelog.txt · /changelog.json.