ORCH + Skyvern — orchestrate parallel browser automation agents with state machine and auto-retry

[oxgeneral]

Hey Skyvern community!

I built ORCH — an open-source CLI runtime for running multiple AI agents as a coordinated team. It pairs well with Skyvern for complex, long-running browser automation workflows.

The problem: Skyvern tasks sometimes fail mid-run. When you're running 10 browser agents in parallel, there's no centralized visibility into what's passing, failing, or stuck.

ORCH solves this with:

• State machine per task: todo → in_progress → review → done
• Auto-retry with configurable backoff for failed Skyvern runs
• TUI dashboard showing all agent activity in real-time
• Dependency chains: agent B starts only after agent A succeeds

# Add Skyvern as shell agents in your team
orch agent add "skyvern-1" --adapter shell \
  --cmd "skyvern run task-page1.json"
orch agent add "skyvern-2" --adapter shell \
  --cmd "skyvern run task-page2.json"

# Batch assign with dependencies
orch task add "Scrape product catalog page 1" --assignee skyvern-1
orch task add "Scrape product catalog page 2" --assignee skyvern-2
orch task add "Synthesize catalog data" --assignee summarizer --depends-on <scrape-tasks>

# Watch all agents in parallel
orch tui

All state stored in .orchestry/ YAML files — zero infrastructure needed.

→ GitHub: https://github.com/oxgeneral/ORCH
→ Install: npm install -g @oxgeneral/orch

Happy to discuss integration patterns for Skyvern workflows!