Case Study

dot-ai

Docs

A personal AI coding profile - rules, workflow, and conventions for AI-assisted development. Think of it like dotfiles, but for how an AI writes code for you.

Open Source Free

Overview

dot-ai is a single-file coding profile that configures AI coding assistants with a strict workflow, output rules, and code conventions. Drop it into any project and the AI follows your rules instead of its defaults.

The Problem

AI coding assistants have opinions. They over-explain, add comments you didn't ask for, refactor code outside the scope of a task, and sprinkle "Sure! Great question!" across every response. Over time this wastes tokens, breaks focus, and produces diffs full of noise. Every developer has different preferences, but the AI defaults are the same for everyone.

What Was Built

A single AGENTS.md file with six sections:

  • 6-step workflow - Plan, Edge Cases, Execute, Check, Edge Cases Again, Summarize. Mandatory for every task, no skipping
  • Output rules - code first, no boilerplate, no sycophantic fluff
  • Code rules - simplest working solution, no over-engineering, no speculative features
  • Naming conventions - self-explanatory names, no comments that should be code
  • Review & debugging rules - state the bug, show the fix, stop. No guessing
  • Formatting rules - plain hyphens and straight quotes, copy-paste safe output

What Made It Interesting

The challenge was making rules specific enough to change behavior but generic enough to work across any project and language. Too vague and the AI ignores them. Too specific and you need a different file per codebase. The balance came from targeting failure modes (verbosity, over-engineering, scope creep) rather than prescribing solutions.

References

The profile was shaped by claude-token-efficient (token-saving output rules) and code-review-graph (structural codebase context before making changes).


View Project All projects →