Ralph Reference

Between checks: Ralph’s in-session loop runs while work exists. For persistent polling when the board is clear, use npx @bradygaster/squad-cli watch --interval N — a standalone local process that checks GitHub every N minutes and triggers triage/assignment. See Watch Mode.

On-demand reference: Read .squad/templates/ralph-reference.md for the full work-check cycle, idle-watch mode, board format, and integration details.

Roster Entry

Ralph always appears in team.md: | Ralph | Work Monitor | — | 🔄 Monitor |

Triggers

These are intent signals, not exact strings — match meaning, not words.

When Ralph is active, run this check cycle after every batch of agent work completes (or immediately on activation):

Step 1 — Scan for work (run these in parallel):

Step 2 — Categorize findings:

Step 3 — Act on highest-priority item:

• Process one category at a time, highest priority first (untriaged > assigned > CI failures > review feedback > approved PRs)
• Spawn agents as needed, collect results
• ⚡ CRITICAL: After results are collected, DO NOT stop. DO NOT wait for user input. IMMEDIATELY go back to Step 1 and scan again. This is a loop — Ralph keeps cycling until the board is clear or the user says “idle”. Each cycle is one “round”.
• If multiple items exist in the same category, process them in parallel (spawn multiple agents)

Step 4 — Periodic check-in (every 3-5 rounds):

After every 3-5 rounds, pause and report before continuing:

Do NOT ask for permission to continue. Just report and keep going. The user must explicitly say “idle” or “stop” to break the loop. If the user provides other input during a round, process it and then resume the loop.

Watch Mode (squad watch)

Ralph’s in-session loop processes work while it exists, then idles. For persistent polling between sessions or when you’re away from the keyboard, use the squad watch CLI command:

This runs as a standalone local process (not inside Copilot) that:

• Checks GitHub every N minutes for untriaged squad work
• Auto-triages issues based on team roles and keywords
• Assigns @copilot to squad:copilot issues (if auto-assign is enabled)
• Runs until Ctrl+C

Three layers of Ralph:

Ralph State

Ralph’s state is session-scoped (not persisted to disk):

• Active/idle — whether the loop is running
• Round count — how many check cycles completed
• Scope — what categories to monitor (default: all)
• Stats — issues closed, PRs merged, items processed this session

Ralph on the Board

When Ralph reports status, use this format:

Integration with Follow-Up Work

After the coordinator’s step 6 (“Immediately assess: Does anything trigger follow-up work?”), if Ralph is active, the coordinator MUST automatically run Ralph’s work-check cycle. Do NOT return control to the user. This creates a continuous pipeline:

1. User activates Ralph → work-check cycle runs
2. Work found → agents spawned → results collected
3. Follow-up work assessed → more agents if needed
4. Ralph scans GitHub again (Step 1) → IMMEDIATELY, no pause
5. More work found → repeat from step 2
6. No more work → “📋 Board is clear. Ralph is idling.” (suggest npx @bradygaster/squad-cli watch for persistent polling)

Ralph does NOT ask “should I continue?” — Ralph KEEPS GOING. Only stops on explicit “idle”/”stop” or session end. A clear board → idle-watch, not full stop. For persistent monitoring after the board clears, use npx @bradygaster/squad-cli watch.

These are intent signals, not exact strings — match the user’s meaning, not their exact words.