How this skill is triggered — by the user, by Claude, or both
Slash command
/plugin-hive:hive-driveThis skill is limited to the following tools:
The summary Claude sees in its skill listing — used to decide when to auto-load this skill
Drive a swarm from start to finish autonomously. This skill enters a wave-based loop: pollinate, spawn bees as background Task agents, monitor via TaskOutput, re-pollinate, and repeat until the swarm completes or fails.
Drive a swarm from start to finish autonomously. This skill enters a wave-based loop: pollinate, spawn bees as background Task agents, monitor via TaskOutput, re-pollinate, and repeat until the swarm completes or fails.
Arguments: $ARGUMENTS
Pattern: /hive-drive <blueprint_id> <task...>
Example: /hive-drive feature-dev Add user authentication system
mcp__hive__hive_swarm_start with the blueprint_id and task.Pattern: /hive-drive <number>
Example: /hive-drive 3
mcp__hive__hive_swarm_summary with the swarm_id resolved from the number.buzzing. If not, report the status and exit.Pattern: /hive-drive
mcp__hive__hive_swarm_list with status=buzzing.The core execution loop. Repeat until the swarm completes, fails, or safety limits are hit.
wave — Counter starting at 1 (max 50)empty_polls — Counter of consecutive empty pollinates (max 3)last_epoch — Last seen epoch value (start at -1)swarm_id — The swarm being drivenStep 1: Check epoch (skip if first wave)
mcp__hive__hive_check_epoch.last_epoch, sleep 5 seconds and re-check (up to 3 times).last_epoch.Step 2: Pollinate
mcp__hive__hive_pollinate with the swarm_id.empty_polls.empty_polls to 0.Step 3: Spawn bees
Task tool to spawn a background agent:
subagent_type: "general-purpose"run_in_background: trueprompt: The SpawnRequest's prompt fieldmodel: The SpawnRequest's model fieldmax_turns: The SpawnRequest's maxTurns field{beeId}: {flightDescription}"Step 4: Monitor background tasks
TaskOutput with block=true.Step 5: Check swarm status
mcp__hive__hive_swarm_summary with the swarm_id.completed: call mcp__hive__hive_swarm_analytics and report success with timing summary. Exit loop.failed: report failure and exit loop.cancelled: report and exit loop.blocked: Check for gated flights. Show the gated flight details and ask the user: "Approve this flight to continue? (flight: {flight_id}, gate: approval)". If the user approves, call mcp__hive__hive_gate_approve with the flight_id. Resume the loop.Step 6: Safety checks
wave >= 50: report max waves reached, suggest /hive-drive {number} to continue. Exit loop.empty_polls >= 3: call mcp__hive__hive_beekeeper_check. If beekeeper finds issues, reset empty_polls to 0. If no issues found, report possible stall and exit loop.Step 7: Next wave
wave.Report progress after each wave:
Wave {N} — Swarm #{number}
Dispatched: {count} bee(s)
Pipeline: {done}/{total} flights done
Cells: {done}/{total} done
Status: {status}
On completion:
Swarm #{number} completed successfully!
Task: {task}
Waves: {count}
Flights: {total} completed
Cells: {done}/{total} done
Total time: {duration}s
Bottleneck: {flight_id} ({duration}s)
Parallelism: {ratio}x
On failure:
Swarm #{number} failed.
Task: {task}
Waves: {count}
Use `/hive resume {number}` to retry failed flights.
hive_flight_claim, hive_flight_complete, or hive_flight_fail directly. Bees call these.hive_check_epoch to avoid expensive polls when nothing has changed./hive-drive {number} to continue.npx claudepluginhub twofoldtech-dakota/plugin-hiveSpawns isolated agents to execute independent working-tree tasks in parallel with wave-validity gating and disjoint file ownership checks.
Coordinates multi-agent swarms with queen-led architecture, Byzantine consensus, and shared persistent memory. Use for orchestrating agent teams under a lead coordinator or building resilient cross-agent collaboration.
Runs the hive experiment loop for autonomous iteration on shared tasks in hive directories. Use to run experiments, submit results, or join agent swarms.