From harness
Use when the user says "/harness:mission", "create a mission", "start a mission", "check mission status", "approve this task", "reject this task", or wants to create, monitor, or manage autonomous multi-task missions from Claude Code.
How this skill is triggered — by the user, by Claude, or both
Slash command
/harness:missionThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
<span data-proof="authored" data-by="ai:claude">Create, monitor, and manage autonomous missions from Claude Code. Missions are multi-task workflows executed by The Harness orchestrator — each task spawns an independent Claude CLI session.</span>
Create, monitor, and manage autonomous missions from Claude Code. Missions are multi-task workflows executed by The Harness orchestrator — each task spawns an independent Claude CLI session.
Call harness_status to verify the daemon is running.
harness_create_mission with the detailsNote: Missions need tasks before they can start. Tasks are typically added via the dashboard's decomposition feature or REST API.
harness_get_mission to get the overviewharness_get_mission_tasks to see all tasks with statusesMission: {title} [{status}]
Tasks: {completed}/{total} ({running} running, {review} in review)
[completed] Task 1: Set up database schema
[running] Task 2: Implement API endpoints
[review] Task 3: Write integration tests ← needs approval
[pending] Task 4: Deploy to staging
When tasks reach review status, they need human approval:
Call harness_get_mission_tasks to find tasks in review state
For each review task, present the task details and ask the user
Based on user decision:
Approve: Call harness_approve_task — task moves to completed, downstream tasks unblock
Reject: Call harness_reject_task with feedback — task resets to pending with incremented retry count, feedback appended to description
Start: Call harness_start_mission — sets status to running, orchestrator begins dispatching tasks
Pause: Call harness_pause_mission — stops new task dispatch, running tasks continue to completion
Call harness_list_missions with optional status filter to see recent missions.
Keep status updates concise — use the text progress view, not verbose descriptions
For review gates, show the task title and any output/artifacts before asking for approval
Don't ask for confirmation on read-only operations (list, get, status)
Do ask for confirmation before start, pause, approve, reject
npx claudepluginhub lukaleskovsek/makers-and-breakers-claude-marketplace --plugin harnessCreates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.