todo.txt

# Coders Plugin - Task List
# Derived from PRD roadmap and improvements

## 🚨 HIGH PRIORITY
[SKIP] **Debug spawn persistence bug** - SKIPPED: This is a Node.js-specific issue. The Go rewrite uses native binaries which don't have the same process lifecycle/signal propagation problems.

## Phase 1: Reliability & Polish
[x] Implement auto-reconnect logic for Redis disconnections
[x] Add health checks for proactive detection of stuck sessions
[x] Implement automatic session restart on crash (error recovery)
[ ] Add structured logging for better debugging
[ ] Add TUI session preview (live preview of selected session in split view)
[x] Create configuration file support (YAML/JSON) for default settings
[x] Add `--ollama` spawn flag to map `CODERS_OLLAMA_*` env vars to `ANTHROPIC_*` for Claude Code sessions (see `OLLAMA_CLAUDE_CODE.md`)
[ ] Implement graceful degradation when Redis is unavailable (local fallback)
[ ] Document minimum tmux version requirements

## Phase 2: Enhanced Coordination
[ ] Implement task dependencies to define prerequisites between agents
[ ] Build shared context system for agents to share discovered information
[ ] Create result aggregation to collect and merge outputs from multiple agents
[ ] Implement event-driven workflows to trigger actions based on session events
[ ] Build agent-to-agent handoff for delegating subtasks

## Phase 3: Developer Experience
[ ] Develop VS Code extension for native IDE integration
[ ] Create session templates for pre-defined multi-agent configurations
[ ] Implement cost tracking to monitor API usage across agents
[ ] Add output archiving to save session transcripts for review
[ ] Build replay mode to re-run sessions from saved state

## Phase 4: Scale & Distribution
[ ] Implement remote sessions via SSH-based agent spawning
[ ] Add container isolation using Docker for session sandboxing
[ ] Create Kubernetes operator for cloud-native multi-agent orchestration
[ ] Build team sharing for session state across team members
[ ] Implement queue-based spawning with rate-limited agent creation

## Phase 5: Intelligence
[ ] Implement smart routing to automatically select best AI for task type
[ ] Build context optimization to compress and prioritize context for agents
[ ] Add learning from history to suggest similar past solutions
[ ] Implement conflict detection to warn when agents might conflict
[ ] Create progress estimation based on patterns

## Risk Mitigations
[ ] Implement version pinning and adapter pattern for AI CLI breaking changes
[ ] Add automatic cleanup with TTLs to prevent session resource leaks
[ ] Implement sandboxing and permission controls for security

## Current Limitations to Address
[ ] Add cross-platform support beyond macOS (iTerm2 integration)
[ ] Implement distributed multi-machine coordination
[ ] Add automatic session restore after reboot
[ ] Create adapters for tool-specific quirks (TUI vs stdin handling)

## Performance Targets
[ ] Optimize session spawn time to under 2 seconds
[ ] Ensure dashboard update latency under 100ms
[ ] Achieve heartbeat reliability above 99.9%
[ ] Improve session restore success rate above 95%