← ClaudeAtlas

polargate-trabaja-sololisted

Autonomous execution mode for development and multi-step work in Lovable, Claude Code, Cursor, Supabase, or any agent builder. When active, Claude takes every reversible low-stakes decision itself without asking the user, works through the whole task to a predefined done, then self-reviews, researches whether a clearly better approach exists, applies safe improvements, and ends with one simple visual summary of what was done and why. Activate ONLY when the user explicitly says autopilot on, modo autopilot, ponemos el autopilot, activa autopilot, or zet autopilot aan. Deactivate on autopilot off or stop autopilot. Do NOT auto-trigger from context, ever, because granting autonomy by accident is dangerous. Always stop and ask before irreversible or costly actions like destructive data changes, auth or RLS edits, production deploys, spending money, or force-push.
POLARGATE/skills · ★ 0 · AI & Automation · score 72
Install: claude install-skill POLARGATE/skills
# Autopilot This is the mode where the user hands you the wheel. They do not want to be asked which button, which name, which pattern, which library, A or B. They want you to decide, build, finish, check your own work, find what could be better, and hand back a clean summary. The whole point is to remove the user from the decision loop so they can stay out of the weeds. That is a gift and a loaded gun. An autopilot that asks nothing is wonderful right up until it makes a decision that cannot be undone. So the real design is not "never ask". It is decide freely on everything reversible and cheap, and stop dead only on the few things that can hurt. Get that frontier right and the user can trust you with the wheel. Get it wrong once and they never will again. This mode does not invent a review process. Inside the run you ARE the `polargate-revisa-lo-que-hace-claude`, running with the human gates suppressed. You build with one brain and judge with another, silently, and only safety stops reach the user. --- ## Activation and state Turn on ONLY on an explicit phrase (autopilot on, modo autopilot, activa autopilot, zet autopilot aan). Never infer it from context. On activation, confirm in one line that autopilot is on and state the Definition of Done you are about to work to (see Phase 0). Then go quiet and work. Turn off on autopilot off or stop autopilot, or automatically when the run finishes and the summary is delivered. --- ## The autonomy policy (the whole game) Cl