Review OpenAI Symphony spec for TaskYou compliance

[bborn]

Summary

• Reviewed the OpenAI Symphony SPEC.md against TaskYou's architecture
• Created detailed comparison analysis in docs/symphony-spec-review.md
• Identified gaps, learnings, and areas where TaskYou already exceeds Symphony

Key Findings

TaskYou already covers most Symphony functionality and goes beyond it in several areas:

• Pluggable AI executors (Claude, Codex, Gemini, etc.) vs Symphony's Codex-only
• Human-in-the-loop completion (agents can't mark tasks as done)
• Rich TUI with Kanban board vs Symphony's optional minimal dashboard
• Git worktree isolation vs plain directories
• MCP server per task with rich tooling
• Session resume, task dependencies, context caching

Top gaps worth closing from Symphony:

1. Concurrency limits - Add max_concurrent_tasks setting (TaskYou currently runs all queued tasks in parallel)
2. Stall detection - Kill/retry tasks with no output for N minutes
3. Auto-retry for transient failures - With exponential backoff for crashes/timeouts
4. Token/usage tracking - Capture and display per-task token consumption

Full Symphony compliance wouldn't make sense - the architectures differ fundamentally (Linear-driven daemon vs developer-oriented TUI tool). A better path would be a "Symphony-compatible mode" adapter layer.

Test plan

• Review the analysis document for accuracy and completeness
• Evaluate prioritized recommendations for roadmap inclusion

🤖 Generated with Claude Code