Poteto Mode: Adaptive Mission Autopilot
Programming
Open PlaybookSet the Mission
Run the Mission in Poteto Mode
Give the grounded mission to Poteto Mode and let it own task classification, pstack playbook selection, rigor, model choice, delegation, parallelism, verification, and stopping behavior.
Do not reproduce pstack's internal routing in this Playbook. Poteto Mode may keep the work focused, use arena or swarm, drive one predicate through an autonomous run, design a bespoke workflow, orchestrate a long-running program, or enter an autopilot flow. Let the mission, its scale, the requested delivery outcome, and the operator's words determine the route.
Autonomy is adaptive unless the operator says otherwise. Proceed through reversible work without repeatedly asking for permission. Do not infer permission for irreversible actions, deployment, destructive writes, customer communication, or merging when that authority was not granted. A request to explain the plan is not an instruction to start an autopilot program.
Include durable-state handling in the Poteto Mode brief. Reuse an existing Case when it already owns the mission. Create a Case when the work would materially benefit from cross-session continuity, multiple owners, asynchronous review, durable decisions, or a reliable handoff. Do not require a Case for a short self-contained run. When a Case is unnecessary, use the closest existing artifact or leave a concise final handoff.
When a Case is used, store only durable state such as important decisions, verified checkpoints, blockers, acceptance, and handoff. Create Tasks only when work needs explicit ownership or review. Do not store raw tool traces, credentials, transient retries, or chain-of-thought.
For long-running or unattended work, use pstack's show-me-your-work skill to maintain a reviewable decision trail and support pause or resume. Continue until the mission is complete, a real human gate is reached, or evidence shows a genuine blocker.
Accept the Outcome
Close the Loop
Leave the mission understandable without the original conversation.
Record the outcome, relevant evidence, important decisions, remaining risks, follow-ups, and artifact references in the state surface used during execution. If a Case exists, append the final durable state there and close it only when the accepted outcome and authorization allow. If no Case is needed, leave the same minimum information in the existing artifact or final handoff.
Create follow-up Tasks only for work that needs separate ownership or review. Do not keep the original mission ambiguously open because unrelated work was discovered.
Look for reusable learning after the outcome is settled. Use pstack's reflect skill only when the run produced a repeatable debugging method, verification technique, workflow correction, tool, or engineering rule that would materially improve future execution. Skip reflection for routine completion or one-off context. Prefer encoding the lesson in a test, script, lint, skill, or structural constraint when possible.