Missions and collaboration

Create a mission when a request needs a durable owner, history, or handoff. Use updates for progress and decisions, and ask a question when an agent cannot safely continue.

Because discussion stays on the mission, another teammate or agent can pick up the work without reconstructing context from chat or terminal logs.

What streams into a mission

While an agent is working, the mission can receive:

• session state updates
• progress summaries
• blocking questions
• follow-up messages
• final delivery events

The board streams these in real time, and missions track read / unread state per reviewer so it is easy to see what needs attention since you last looked.

Human-in-the-loop questions

When an agent is blocked on a decision, it publishes a question to the mission and stops working the blocked path. You can respond directly in the mission and keep the work moving without losing context. Answer the question before restarting the run.

Ask when a decision would change the shape of the delivery — an API surface, a data model, or a migration direction — and the objective does not specify it. When the objective or prior context already answers a question, the agent decides and records the reasoning in its next update.

Shared context that survives sessions

Missions support a persistent, mission-scoped shared context so useful facts survive across sessions. It is a small key/value store for durable facts — architecture notes, deploy targets, the path to environment secrets — not conversation history.

# An agent writes a fact for future sessions
ovld protocol write-context --session-key <key> --mission-id <mission-id> \
  --key "deploy.target" --value '{"provider":"vercel","project":"overlord-web"}'

# And reads it back later
ovld protocol read-context --session-key <key> --mission-id <mission-id>

That helps when work pauses, resumes later, or moves between different agent runtimes. See Context and artifacts for the full picture of what agents read and write.

Handoffs

The mission record makes handoffs straightforward: start the next objective with a different agent, and it receives the mission context — original goal, decisions, artifacts, and prior delivery — instead of starting cold.

Next: Review deliveries