Python CLI tool for hierarchical multi-agent communication.
SQLite-backed (WAL mode), agent identity via BTMSG_AGENT_ID env var.
Features:
- inbox/read/send/reply — message CRUD with read tracking
- contacts — role-based communication hierarchy enforcement
- history — per-agent conversation view
- status — all agents with tier/role/model/unread counts
- register/allow — agent and contact management
- notify — single-line notification for agent injection
- Short ID prefix matching for convenience
Also: change default Claude model to opus-4-6
- Add doc/ alongside docs/ in markdown file discovery (groups.rs)
- Add SETUP.md to priority root files
- Fix SSH terminal tabs: resolve session args via sshArgsCache derived
from listSshSessions() instead of passing empty args
- Fix Agent Preview: only mount xterm when tab is active (prevents
CanvasAddon crash on hidden elements)
- Separate tab type rendering (shell/ssh/agent-preview) with proper guards
Remove 500-char assistant text truncation in anchor serializer — research
consensus (JetBrains NeurIPS 2025, SWE-agent, OpenDev ACC) is that agent
reasoning must never be truncated; only tool outputs get observation-masked.
Add AnchorBudgetScale type with 4 presets (Small=2K, Medium=6K, Large=12K,
Full=20K) and per-project range slider in SettingsTab. Remove Ollama-specific
warning toast — budget slider handles context limits generically.
Add Providers section to SettingsTab with collapsible per-provider config
panels and per-project provider dropdown. Implement provider-based sidecar
runner selection and multi-provider env var stripping in Rust.
Worker copied to public/ via prebuild script (avoids Vite resolution issues).
IntersectionObserver renders only visible pages (+200px ahead) instead of all
at once, fixing performance for large PDFs.
pdfjs-dist for canvas-based multi-page PDF rendering with zoom controls.
RFC 4180 CSV parser with delimiter auto-detection and sortable columns.
FilesTab routes Binary+pdf to PdfViewer, Text+csv to CsvTable.
