[ docs · concepts ]
Specialists
A Grist run doesn't do everything itself. It delegates to named specialist subagents — one to investigate, one to design, one to check the diff, one to run the tests — each working inside the rung the gate assigned.
The roster
| Specialist | Job |
|---|---|
explore | Read-only investigation: finds the relevant files, symbols, and history before anyone edits. |
plan | Designs the change: what moves, in what order, and what could break. |
review | Checks diffs against the intent — the second pair of eyes before anything lands. |
verify | Runs the project's checks and tests, and reports what actually passed. |
Delegations re-gate
Spawning a specialist is not a blank cheque. Each delegation is scored again, on its own text, by the Jev gate — with the parent's rung as a ceiling. A frontier run can hand a mechanical sub-task to the cheapest rung; a cheapest run can never hand anything upward. Decomposed work can only shed cost, never gain it.
Warm resumes
Calling the same specialist twice doesn't start it twice. Repeat calls resume warm: the specialist keeps the context it already built — the files it read, the map it formed — and picks up where it left off. You pay for the investigation once, not once per question.
Specialists are how a run stays honest: the one who writes the change is never the only one who checks it.