I-have-ADHD: A skill to stop coding agents from burying the answer

TL;DR Highlight

A 10-rule skill makes coding agents lead with actions and numbered steps

Who Should Read

A developer tired of Claude Code burying answers under preambles and closers

Core Mechanics

It puts the answer first
- It stops coding agents from burying the answer.
- Action first, numbered steps, and no "Hope this helps!".
Before/After shows the difference
- Before starts with "Great question!", explains the auth flow at length, and ends with "Hope this helps!".
- After lists numbered steps: edit src/auth.ts:42, replace verifyToken (lines 42–58), run npm test -- auth.spec.ts.
- It ends with one next step: "Next: paste the first failing line".
There are 10 rules
- Lead with the next action, number multi-step tasks, and end with one concrete next step.
- Suppress tangents and restate state every turn.
- Give time estimates in minutes, make wins visible, and write errors matter-of-factly.
- Cap lists at 5 items and drop preambles, recaps, and closers.
Installation is delegated to the agent
- Paste a prompt with the repo URL telling the agent to follow AGENTS.md.

Community Reactions

Report that skills wear off quickly
- One user said Claude gets verbose again within a few turns.
- This happened even with the instruction in CLAUDE.md and manual reminders.
- Another used it for months and found output styles work better, though only one can be active at a time.
Opinion that config or a plain instruction is enough
- One asked why this is a skill and not AGENTS.md/CLAUDE.md config.
- Another said this is a user-level preference, neither repo context nor a task skill.
- Another said just tell it to use BLUF and ASD-STE100.
Alternative experiences for keeping style
- One user gave Claude a response function with a required style parameter and got tighter code.
- Another lints output with a vale script and asks for plain technical English.
Concerns about security and size
- Some worried that a one-line install prompt could end up at a similarly named repo.
- One noted the prompt is 140 lines while the repo has 59 files and 8.7k lines.

How to Apply

Put the core 5 rules in CLAUDE.md
- Today, add "action first, numbered steps, no preamble/recap/closer, end with one next step" to CLAUDE.md.
- Compare before and after answers on 3 of your usual questions.
Re-state the rules in long sessions
- When replies get wordy after a few turns, paste the rule in one line again.
- Compare whether the skill or an output style lasts longer.
Fork SKILL.md into your team's rules
- Copy SKILL.md and add bans for phrasing that annoys your team.
- Reinstall and verify with /i-have-adhd.
Check the repo URL and AGENTS.md before installing
- Before telling an agent to install, read the URL and AGENTS.md yourself.
- After installing, check which files were added.

Try It

copy & run

# Paste at the start of a Claude Code or ChatGPT chat (simplified version of the repo's rules)

从现在起 follow these rules for every answer.
1. Start with the next action.
2. Number multi-step tasks.
3. End with one concrete next step.
4. No tangents.
5. Restate the current state in one line each turn.
6. Give time estimates in minutes.
7. Mark finished work visibly.
8. State errors matter-of-factly.
9. Cap lists at 5 items.
10. No preamble, recap, or closers like "Hope this helps".

Question: {{question}}

Try next: ask the same question without the rules and compare the two answers

Terminology

skillA bundle of instructions telling a coding agent how to do a certain kind of work. Like handing the agent an onboarding doc.

CLAUDE.md / AGENTS.mdProject instruction files the agent reads at session start. Like a team wiki's must-read page.

output styleA Claude Code setting that defines response tone. It reminds the model how to speak on each response.

BLUFBottom Line Up Front: put the conclusion first. Like a headline before the article body.