As An Engineer offers four heuristics for AI coding agents

Summary

As An Engineer is a repository-hosted, heuristic skill intended to make AI coding agents treat users as senior engineers rather than defaulting to what its author calls “babysitting.” It frames that behavior as a source of engineering anti-patterns and sets out four checks to apply before choosing an approach. First, when the upstream format, schema, headers, or metadata already carries the needed signal, extract it instead of deriving it through heavier computation. Second, for high-friction or credential-heavy work, provide a runnable artifact for the user’s environment rather than trying to execute the task in a restricted agent sandbox. Third, keep data separate from presentation, preserve composable boundaries, and expose operational limits as configurable inputs instead of hardcoding them. Fourth, when candidate paths differ greatly in cost, ask a short question to verify the prerequisite for the cheaper path rather than guessing and launching a heavy process. The project provides installation instructions for Antigravity or the Agent Skill standard, Claude Code, Cursor, and generic agent rules. It is released under the MIT license.