How a run works
A scheduled run, step by step: it starts, reads context and numbers, reports a finding or stays quiet and finishes.
Updated
A run is one time a skill runs on its schedule. Each step below is a tool call your Claude makes on our MCP server, so you can follow a run the way your agent sees it.
1. Start
The routine calls start_run. It returns the next run that's due: which skill, which project and which tools the run may use. With nothing due, the run stops there.
2. Read the context
get_context gives the project's facts and your team's project knowledge: instructions, memory, goals. The skill's own instructions come with get_skill.
3. Read the numbers
The skill's instructions decide what to read: get_dashboard, query, find_changes, get_timeline, list_insights.
4. Report, or stay quiet
When it finds something worth your attention, the run calls post_finding once: a title, a short summary, the numbers and the query that proves them. When there's nothing worth saying, it posts nothing. A quiet run is a good run.
5. Remember
suggest_memory can leave a short note in the team's memory: a fact or what worked, with its numbers.
6. Finish
finish_run ends the run, done or failed with the reason. Then the routine asks start_run again, until nothing's left.