I Have ADHD
An agent skill that stops Claude Code from burying the answer. Ten rules live in a single markdown file: lead with the next action, number multi-step work, close on exactly one concrete step, suppress tangents, restate state each turn, give real time estimates, make wins visible, report errors flatly, cap lists at five items, and drop preamble, recap and closing pleasantries. The before and after is blunt. Without it, an answer opens by praising the question, restates the problem, volunteers options nobody asked for, and hides the useful command on the last line. With it, the command comes first, then the file to edit, the package to install, the test to run. Installs through the Claude plugin marketplace and is invoked on demand. The repository also ships config files for Codex, Cursor, Gemini, Kimi and Qwen. Everything is markdown, so forking it and rewriting the ten rules to your own taste takes minutes.
Strengths
- The useful command lands on the first line, which removes the visual scan on every answer
- Everything fits in a ten-rule markdown: you can read it whole, fork it and rewrite it
- The repository ships configs for Codex, Cursor, Gemini, Kimi and Qwen, not just Claude Code
Limitations
- The five-item cap also truncates the inventories you actually wanted complete
- Without a recap, you lose the moment where you used to catch the agent misreading the request
- Three same-named repositories circulate under close names, you have to check which one you install
Best for
- A debugging session where you chain twenty round trips and every preamble is paid for in reading
- A profile that drops off on long answers and wants the action before the justification
- Working on a narrow screen, a terminal or a phone, where three lines of politeness cost a scroll