Skip to content

TODO.md

Latest commit

 

History

History
72 lines (52 loc) · 3.61 KB

TODO.md

File metadata and controls

72 lines (52 loc) · 3.61 KB

AI Office — TODO

Keep this file up to date. When completing work on any item below, update its status immediately.

Agent Observability

  • Click agent cards on manager kanban to see their session (running or done)
  • Log streaming when clicking on working agents to see what their computer looks like
  • Partial Observability into computer use of workers — see their screen, where they click, how they navigate (read-only). JPEG screenshots work; missing live VNC stream and click/navigation overlay annotations.
  • Expandable thoughts-type message + then a proper message

Agent Capabilities

  • Improve code agent capabilities (Mistral Vibe headless, auto-verify, auto-commit on feature branches)
  • Daytona sandbox scaffolding fixes (PTY for interactive CLIs, local workspace for npm install, coder snapshot with Vibe pre-installed)
  • Give more Daytona tools to agents depending on agent type (role-based capability tiers in capabilities.ts)
  • Git operations — setup login to access git user credentials for agents (full git toolkit, GitHub token injection)
  • Comment on the repo (createIssue, addComment, createPR via gh CLI)
  • Improve system prompting (role-based prompts, two-phase internal dialog, no-fabricated-links rule)
  • Plan + execute mode (researcher: plan/execute/synthesize, general: plan/retry/summary, browser: iterative vision loop)
  • Partial Full computer access, talking to the manager directly. Computer use implemented but disabled on Daytona Tier 1/2 (no outbound internet). Manager communication works via mailbox.

Task Management

  • Task board with dependencies
  • Dependency system
  • Per-worker kanban in sidebar
  • What the task is, task threads

Agent Orchestration

  • Automatic agent orchestration system like OpenClaw (manager decomposes, spawns, monitors, handles completion)
  • Partial Creates custom agents with defined skill + personality, can delegate skills/tasks. 6 pre-defined roles exist but no dynamic agent creation UI.
  • Tools: spawn, steer (message), stop agent
  • Worker complete sends messages to the main agent
  • Message queuing (priority-based mailbox: critical/high/normal/background)
  • Partial Clock in / clock out with work duration. spawnedAt/completedAt timestamps tracked, but no explicit clock-in/out UI.

Multi-Project System

  • Project system — different projects with:
    • Dedicated Daytona volume
    • Dedicated manager thread
    • Dedicated task kanban
    • Separate deliverables
  • Partial Multiple chat threads, allow for multiple projects setup. Multiple chat threads exist but not project-scoped.
  • + Add Room, reorder rooms, change floors as a funny thing

Deployment

  • Deploy the project somewhere (maybe AWS) (GitHub Actions → S3 bucket, auto-deploys on push to main)
  • Maybe Vercel deploy via API — making a deployed project
  • Maybe deploy in an EC2 VPS

User Interaction

  • User prompting (chat UI at /ai, Telegram bot, structured Q&A system)
  • Documents location — file icons on desk, documentation and plans there (Document Hub at /docs)

UI / Fun

  • Mistral cat (kitchen corner sprite)
  • Speech bubbles (permission + waiting bubbles above agent heads)
  • Partial Easter eggs. Mistral cat is one, no others found.
  • Linear integration — task management like Linear, link Linear, assign agents

Queue & Notifications

  • Queue for messages (full mailbox system with priority queue)
  • Comment-to-task agent notification (auto-notifies assigned agent or manager via mailbox)