Menú
Use when raw intent must be captured and routed to the right next state.
Instalar con Codex o Claude Copia este prompt, pégalo en Codex, Claude u otro asistente, y deja que revise la página de la skill y la instale por ti.
Basado en la clasificación ocupacional SOC
Agent Brain /brain-brief: after intake, research, and grill have enough signal.
Agent Brain /brain-build: an implementation plan has a selected task and validation method.
Agent Brain /brain-design: a product brief needs ux or interaction design before planning.
Agent Brain /brain-eval: changing Agent Brain behavior or checking quality.
Agent Brain /brain-grill: the idea, brief, design, or plan has unresolved assumptions.
Agent Brain /brain-learn: after repeated success/failure, a tricky fix, or a shipped workflow.
| name | intake |
| description | Use when raw intent must be captured and routed to the right next state. |
Lifecycle stage: INTAKE
Use when a request is vague, new, or missing context.
Do not use when this trigger is absent; choose the command or skill that owns the requested state, artifact, and verification gate.
| Shortcut | Rebuttal |
|---|---|
| "This is obvious." | Write the assumption and evidence. If you cannot, it is not obvious. |
| "We can do this later." | If the missing step changes the decision, do it now or state the risk. |
| "The user wants speed." | Reduce scope; do not skip the quality bar. |
| "This does not need verification." | Every important claim or behavior needs proof. |
Intake Summary
Use templates/intake-summary.md. The artifact should be concise, auditable, and include evidence, blockers, and next action so another agent can resume.
Trigger: raw request enters the harness. Action: capture facts, assumptions, constraints, missing evidence, approval state, and likely next command before deeper work. Output artifact: templates/intake-summary.md with blockers and next action. Verification: cite routing evidence and the safest noninteractive default when follow-up is impossible.
A user says, "Make the harness better." Intake should capture known facts, assumptions, constraints, missing evidence, approval state, and likely next command; in a scheduled run, choose the safest documented default and create an intake summary rather than asking a broad follow-up.
