h2-qa.yaml

sandbox:
  dockerfile: qa/Dockerfile
  setup:
    - "h2 init ~/.h2"
  env:
    - ANTHROPIC_API_KEY

orchestrator:
  model: opus
  extra_instructions: |
    You are testing h2, a terminal multiplexer with inter-agent messaging.

    ## Available h2 commands
    - h2 run <role> — Start an agent with a role
    - h2 run --dry-run <role> — Show resolved config without launching
    - h2 pod launch <pod> — Launch a pod (group of agents)
    - h2 pod launch --dry-run <pod> — Show pod config without launching
    - h2 stop <agent> — Stop a running agent
    - h2 pod stop <pod> — Stop all agents in a pod
    - h2 list — List running agents
    - h2 peek <agent> — View an agent's terminal output
    - h2 send <agent> "message" — Send a message to an agent
    - h2 init <dir> — Initialize an h2 config directory

    ## How messaging works
    - Messages are delivered via files in ~/.h2/messages/<recipient>/
    - Agents see messages prefixed with [h2 message from: <sender>]
    - Priority levels: interrupt (immediate), normal (next pause), idle (when idle)
    - Use h2 send for all inter-agent communication

    ## Verifying agent state
    - h2 list shows agent name, PID, status, and role
    - h2 peek <agent> shows recent terminal output (useful for checking if agent responded)
    - Check ~/.h2/sessions/<agent>/ for session files
    - Check exit codes and stderr for error conditions

    ## Test agent roles
    - Use haiku or sonnet models for test sub-agents to minimize cost
    - Create minimal roles in ~/.h2/roles/ as needed for each test case
    - Clean up roles and agents after each test case

plans_dir: qa/plans/
results_dir: qa/results/