BD Swarm Auto - Autonomous Backlog Completion

YOU are the conductor. Execute this workflow autonomously. Do NOT ask the user for input.

Execute Now

1. Check ready issues:

   bd ready --json | jq -r '.[] | "\(.id): \(.title)"'
   

   If empty, announce "Backlog complete!" and stop.

2. Create worktrees in parallel:

   PROJECT_DIR=$(pwd)
   WORKTREE_DIR="${PROJECT_DIR}-worktrees"
   mkdir -p "$WORKTREE_DIR"
   
   for ISSUE_ID in $(bd ready --json | jq -r '.[].id'); do
     ${CLAUDE_PLUGIN_ROOT}/scripts/setup-worktree.sh "$ISSUE_ID" &
   done
   wait
   

3. Spawn workers (max 4):

   TOKEN=$(cat /tmp/tabz-auth-token)
   for ISSUE_ID in $(bd ready --json | jq -r '.[].id' | head -4); do
     WORKTREE="${WORKTREE_DIR}/${ISSUE_ID}"
     RESPONSE=$(curl -s -X POST http://localhost:8129/api/spawn \
       -H "Content-Type: application/json" \
       -H "X-Auth-Token: $TOKEN" \
       -d "{\"name\": \"worker-$ISSUE_ID\", \"workingDir\": \"$WORKTREE\", \"command\": \"claude --dangerously-skip-permissions\"}")
     SESSION=$(echo "$RESPONSE" | jq -r '.terminal.ptyInfo.tmuxSession')
     echo "Spawned $ISSUE_ID -> $SESSION"
   done
   

4. Send skill-aware prompts: For each worker, send a prompt with:

   • Issue context from bd show
   • Skill hint based on issue keywords (terminal→/xterm-js:xterm-js, UI→/ui-styling:ui-styling, etc.)
   • Completion command: /conductor:worker-done <issue-id>

5. Monitor and loop:

   • Poll ${CLAUDE_PLUGIN_ROOT}/scripts/monitor-workers.sh --summary every 2 min
   • When all issues closed, run completion pipeline
   • Check bd ready - if more issues, start next wave

Key Rules

• NO USER INPUT - Make reasonable defaults, never AskUserQuestion
• MAX 4 TERMINALS - Distribute issues across workers
• MONITOR CONTEXT - At 70%+, trigger /wipe:wipe with handoff
• LOOP UNTIL EMPTY - Keep running waves until backlog clear

Full Details

See skill: /conductor:bd-swarm-auto for complete workflow reference.