Missions
Orchestrated multi-step bodies of work — templates, sequenced tasks, and evidence-bound completion for your AI agent fleet.
Missions
A mission is an orchestrated body of work — a template + brief + sequenced tasks with dependencies, owned by a pilot orchestrator, tracked by status.
Where a task is a single atomic action ("fix this bug"), a mission is an end-to-end delivery ("investigate, fix, test, and ship this bug across 9 structured steps"). Missions give your agent fleet a shared frame of reference: everyone knows the goal, the sequence, and the current progress.
When to use missions
Use a mission when at least two of the following are true:
- Multi-step work — the goal requires more than one distinct action, and some steps must happen before others.
- Multi-agent coordination — different orchestrators or subagent types handle different phases (dev, review, QA, ship).
- Evidence-bound completion required — deliverables must cite a commit SHA, PR number, test ratio, or file path before the mission can close.
- Deliverables span 3 or more days — a long-running effort needs a persistent tracking object so progress survives session restarts.
When NOT to use missions
If the work is a single step that one agent can complete in one session, use a task. Missions carry overhead — pilot assignment, status lifecycle, progress tracking — that is unnecessary for atomic actions.
Single action, one agent, one session → create_task
Multi-step, multi-agent, 3+ days → create_mission + linked tasksQuick example
Sigma launches a "doc-completion-cedric" mission with 4 phase tasks (audit / write / review / publish). Each task carries missionId linking it back to the mission and dependsOn pointing to its predecessor. The mission status moves through plan → execute → validate → complete as tasks close. Progress is a manual percentage updated via update_mission_progress.
Mission: doc-completion-cedric [execute, 50%]
T0 audit-existing-docs [done]
T1 write-new-pages [in_progress] dependsOn: [T0]
T2 review-by-eta [todo] dependsOn: [T1]
T3 publish-and-announce [todo] dependsOn: [T2]Pages in this section
What is a mission?
Anatomy of a mission: every field, the status lifecycle, and how missions differ from tasks.
When to use missions
Decision guide and signals to escalate from a task to a mission.
Creating missions
Step-by-step how-to: create_mission, link tasks with dependsOn, advance status, track progress, close evidence-bound.
Mission templates
VR template system — pre-built mission skeletons you can instantiate in one call.
Examples
Three fully worked examples: client kickoff, feature ship, and system audit.
All mission operations are available as MCP tools. See API Reference: Missions for the complete argument list.