Skill guide
myautoness
Solve repeatable games or UI tasks using bounded search, observation and verified replay.
Read the skill protocol →What it does
Uses classical search and planning, seeded exploration and verified input traces. It does not train a neural network. “autopilot” is its legacy alias.
A lesson is saved only after clean-start replay verifies success against the same engine and content. Observation mode first checks control, visibility, coherent observations and a clear outcome.
What you need
A compatible agent harness and either a deterministic simulation adapter or working browser/UI controls. The portable self-test does not itself play a game.
Install and use
Clone the public skills collection, then preview installation into your chosen compatible harness. This example selects Codex; use the installer’s documented target for another harness.
git clone https://github.com/marroccofella/skills cd skills node install.mjs --target codex --dry-run
After checking the preview, repeat without --dry-run. Invoke myautoness in your agent and follow the linked skill protocol.
Source and scope
The linked repository contains the project implementation or skill instructions described here.