Add Ralph — built-in work monitor squad member

[spboyer]

Closes #14

Summary

Adds Ralph as a built-in squad member (like Scribe) whose job is keeping tabs on work. Once activated, Ralph self-chains the coordinator's work loop — checking for open issues, draft PRs, review feedback, CI failures — and keeps the squad working through the backlog without manual nudges.

Ralph never stops on his own while work remains. He keeps cycling until every issue is closed, every PR is merged, and CI is green. The only things that stop Ralph: the board is clear, the user says "idle", or the session ends.

What's new

🔄 Ralph — Work Monitor

• Built-in squad member, always on the roster, exempt from universe casting
• Badge: 🔄 Monitor (alongside 📋 Silent for Scribe)
• Self-chaining work loop: after every agent batch, Ralph checks GitHub for more work and immediately continues without asking permission
• Periodic check-in every 3-5 rounds with summary (reports status, keeps going)
• Configurable scope: issues only, skip CI, watch everything

Critical loop behavior

Ralph's loop is explicit and non-negotiable in the coordinator instructions:

• MUST NOT stop and wait for user input between work items
• DO NOT stop. DO NOT wait for user input. IMMEDIATELY go back to Step 1
• Check-ins report progress but say "Continuing..." — they do NOT ask permission
• User must explicitly say "idle" or "stop" to break the loop

Trigger phrases

Action	Phrases
Start	"Ralph, go", "Keep working", "Work until done"
Stop	"Ralph, idle", "Take a break", "Stop monitoring"
Status	"Ralph, status", "What's on the board?"
Scope	"Ralph, just issues", "Ralph, skip CI"

Work categories monitored

Category	Signal	Action
Untriaged issues	squad label, no member sub-label	Lead triages and assigns
Assigned issues	squad:{member} label, unstarted	Spawn agent to pick it up
Draft PRs	Squad member PR in draft	Check if stalled, nudge
Review feedback	Changes requested	Route to author agent
CI failures	PR checks failing	Notify agent to fix
Approved PRs	Ready to merge	Merge and close issue

Heartbeat workflow (between sessions)

squad-heartbeat.yml — runs on cron (every 30 min) + event triggers:

• issues.closed — check for next item in backlog
• pull_request.closed — check for follow-up work
• issues.labeled — triage new squad-labeled issues
• workflow_dispatch — manual trigger

The heartbeat auto-triages issues based on team roles and keywords, and assigns copilot-swe-agent[bot] for @copilot issues (if auto-assign is enabled with COPILOT_ASSIGN_TOKEN).

Files changed

File	Change
.github/agents/squad.agent.md	Ralph section with CRITICAL loop behavior, routing signals, casting exemption, follow-up work hook
templates/roster.md	Ralph row: | Ralph | Work Monitor | — | 🔄 Monitor |
templates/workflows/squad-heartbeat.yml	New — heartbeat workflow with auto-triage and @copilot assignment
index.js	Fix upgrade early-exit path to refresh workflows and agent.md
test/index.test.js	10 new tests for Ralph
docs/features/ralph.md	New — full feature documentation
README.md	Ralph in team table, architecture diagram, What's New

Upgrade safety

• Upgrade does NOT touch .ai-team/ — team state is preserved
• Ralph appears on the roster template; existing teams get Ralph on next init or by adding the row manually
• Heartbeat workflow is squad-owned and refreshed on upgrade

Bug fixed

Upgrade early-exit: When version matches (isAlreadyCurrent), the upgrade path now refreshes workflows and squad.agent.md instead of exiting without updating them.

Testing

• 43 tests passing (10 new for Ralph)
• Tested on spboyer/waza — identified and fixed issue where Ralph stopped after Round 1 (loop instructions were too passive)