i-have-adhd: A Coding-Agent Skill That Forces Action-First Answers

What problem it actually solves

The README is blunt about the pain point: coding assistants bury the answer. A typical response opens with 'Great question!', walks through the auth flow, mentions a file and line numbers, then ends with 'Hope this helps!' The person who asked has to parse paragraphs to find the one command they need. i-have-adhd is a skill for coding agents that rewrites that pattern. It is aimed at anyone who finds long assistant replies hard to follow, whether or not they have an ADHD diagnosis. The tagline says 'No ADHD diagnosis needed.' The target user is a developer who wants the next action first, numbered steps, and no social filler. It does not add features to the agent. It changes the style of the output.

The mechanism: a prompt pack, not a plugin with logic

The repository contains a skill definition, and the core is a single file: skills/i-have-adhd/SKILL.md. That file holds the 10 rules that the agent is meant to follow. The README lists them: lead with the next action, number multi-step tasks, end with one concrete next step, suppress tangents, restate state every turn, give specific time estimates, make wins visible, state errors matter-of-factly, cap lists at 5 items, and no preamble, recap, or closers. The rules themselves are text, though the repository also contains an extensions/i-have-adhd.ts entry plus hooks, scripts and evals directories. It is a text prompt that the Claude Code plugin system loads. The plugin mechanism is what installs the skill and makes it available as a slash command. The before-and-after example in the README shows the intended effect: the 'After' column gives a runnable command, a file path with line numbers, three numbered steps, and a next step. The 'Before' column is a paragraph of context. The whole thing is a behavioral contract written for an LLM.

Getting it running: install, fork, and tune

Installation is a copy-paste prompt. You tell your CLI: 'Install the i-have-adhd skill/plugin from https://github.com/ayghri/i-have-adhd, refer to the repo's AGENTS.md for instructions.' There is also an INSTALL.md for the full steps. The README shows the tuning path for people who want to change the rules. You fork the repo, edit skills/i-have-adhd/SKILL.md, then swap your copy in with four commands: claude plugin uninstall i-have-adhd, claude plugin marketplace remove i-have-adhd, claude plugin marketplace add <your-username>/i-have-adhd, and claude plugin install i-have-adhd@i-have-adhd. The comment in the README explains why you must remove the upstream copy first: the fork and the upstream share both names. After that, restart your coding assistant and re-invoke /i-have-adhd. This is a concrete, testable flow, and the fork-and-swap path is the intended way to customize.

The 10 rules: what they force and what they forbid

Each rule is a direct instruction to the model. Rule 1, 'Lead with the next action,' means the first sentence of a reply should be a command or an edit instruction, not context. Rule 2 demands numbered steps for multi-step tasks. Rule 3 says end with one concrete next step, which could be 'paste the first failing line if any test fails.' Rule 4 suppresses tangents, so the agent should not mention dependency versions or unrelated improvements. Rule 5 is 'Restate state every turn,' which seems to contradict rule 10's 'no recap' until you read it as: repeat the current state of the task, but do not re-explain the whole history. Rule 6 asks for specific time estimates in minutes, not 'a bit.' Rule 7 makes wins visible, which likely means acknowledging completed steps. Rule 8 says state errors matter-of-factly, no apology spirals. Rule 9 caps lists at 5 items, which forces prioritization. Rule 10 bans preamble, recap, and closers like 'Hope this helps.' The combination is strict. A model that follows all 10 will produce terse, action-oriented replies, but it may also feel robotic to users who like a little context.

A real limitation: it fights with conversational workflows

The rules are designed for a specific interaction: a user asks a question, the agent gives a direct answer. That works for debugging, code generation, or explaining a function. It breaks down when the user wants a discussion, a comparison of options, or a tutorial. Rule 10 forbids recap and closers, but a long explanation of a design trade-off is not preamble; it is the content. The 'cap lists at 5 items' rule also forces the agent to omit details that might be necessary in a complex answer. The README's example is a simple auth fix. For a multi-file refactor or an architectural question, the numbered steps could become so compressed that they lose the reasoning behind them. Another point to check is agent support: the tuning commands in the README use the claude plugin CLI, but the repository also carries files for other agents, such as .codex-plugin, .cursor, gemini-extension.json, qwen-extension.json, kimi.plugin.json and .opencode, so check INSTALL.md for your agent.

The alternative: writing your own system prompt or using a custom agent instruction

The obvious alternative is to skip the plugin and put these rules directly into your agent's system prompt or a project-level AGENTS.md file. Many coding agents read a project-level instruction file such as AGENTS.md, and you can add a line like 'Always lead with the next action and number steps.' That gives you the same effect without the plugin dependency. The difference is that i-have-adhd is a self-contained, versioned skill that you can install, uninstall, and fork. A manual prompt is easier to lose or forget to update. The skill also comes with a ready-made set of ten rules, which is the value: you get a prompt pack, not a blank slate. But if you only need one or two rules, writing them into your own AGENTS.md is lighter and does not require a fork. The trade-off is maintenance. The skill's rules are opinionated and may not fit every project, so the fork-and-edit path is the way to adapt them. That is more work than a one-line prompt, but it gives you a structured file to review.

Maintenance, license, and what to check before adopting

The repository is MIT licensed, so you can copy, modify, and redistribute the skill freely, as long as you keep the license notice. The latest release is v0.4.1, published 2026-10-19, which is also the date of the last push. If the Claude Code plugin format changes, the skill may need updates. Before you rely on it, check that the plugin commands still match the current Claude Code version. The tuning path is explicit: fork, edit SKILL.md, swap your copy. That is a real cost, because you must keep your fork in sync with upstream changes if you want them. The README says the rules are loosely based on a book, The Adult ADHD Tool Kit, but the adaptation is for LLM responses, not human habits. That is a useful boundary: the skill is about output format, not about diagnosing or treating ADHD.

Editorial conclusion

Adopt i-have-adhd if you use a coding agent and find yourself scrolling past greetings and context to reach the actual command. Its rules can be tuned by editing one file. Skip it if you prefer conversational explanations; the tuning commands in the README use the claude plugin CLI. Before relying on it, verify that the 10 rules fit your team's style, because the 'no recap' and 'restate state every turn' rules can fight with a workflow that expects full context in every answer. Fork the repo, edit SKILL.md, and test on a real task to see if the trade-off is worth it.

Frequently asked questions

Official sources

1. Official README
2. Project repository
3. README
4. Releases