A GitHub repo called i-have-adhd surfaced on Hacker News with 128 points and 91 comments. It provides a reusable skill that stops coding agents from burying final answers inside long reasoning traces.
Skill: i-have-adhd | Type: System prompt + workflow rule | Effect: Answer-first output | Tested on: Cursor, Claude Code, Aider | License: MIT
What the Skill Actually Does
The skill adds a persistent instruction that forces the agent to state the concrete answer or code change in the first paragraph. All reasoning, alternatives, and caveats must follow afterward.
It works by redefining the agent's output contract: lead with the deliverable, then explain. This matches the exact complaint raised in the HN thread where users reported agents producing 400-800 token responses before revealing the fix.
How to Install and Use It
Clone the repo and copy the system prompt into your agent's custom instructions file. For Cursor, paste the content into Settings > Custom Instructions. For Aider, add it via the --system flag or a .aider.conf file.
Run one test prompt after setup: "Refactor this function to handle null inputs." The agent must now open with the revised code block instead of a multi-paragraph analysis.
"Full setup commands"
aider --system-prompt "$(cat i-have-adhd.md)"
Benchmarks from Early Users
HN commenters reported average response length dropped from 620 tokens to 180 tokens on the same tasks. Time-to-first-usable-answer fell by roughly 65% across 12 reported sessions.
The skill adds zero tokens to input context after the initial system prompt. No extra API calls or model changes are required.
Pros and Cons
- Forces direct answers on every turn
- Works across multiple coding agents without retraining
- Zero cost beyond the one-time prompt copy
- Can feel abrupt for exploratory or architectural discussions
- Requires one extra rule when the user explicitly wants step-by-step reasoning first
How It Compares to Other Prompting Methods
| Technique | Answer Position | Avg Tokens Before Answer | Works in Agents |
|---|---|---|---|
| Chain-of-Thought | End | 450+ | Yes |
| ReAct | Middle | 300+ | Partial |
| i-have-adhd | First paragraph | <50 | Yes |
| Standard system prompt | Variable | 200+ | Yes |
Existing techniques like explicit "think step by step" instructions increase reasoning length. This skill reverses that default.
Who Should Use This Skill
Developers who spend most of their time in Cursor, Windsurf, or Aider and want shorter, actionable replies will see immediate gains. Researchers running long agent sessions for code exploration may prefer to keep full reasoning visible and should skip it.
Teams already using strict output schemas can achieve similar results with JSON mode, but the skill requires no schema changes.
Bottom Line
The i-have-adhd skill delivers the first reliable, zero-cost method to enforce answer-first behavior in coding agents across multiple platforms.
Top comments (0)