6700 Stars for a Skill That Just Tells Claude to Get to the Point
i-have-adhd added 1846 stars in one day and sits at 6.7k. It contains no code that does anything. It is a skill file with ten rules about how the agent should talk, and that is the entire product.
The rules: 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 in minutes, not "a bit". Make wins visible. Report errors matter-of-factly. Cap lists at five items. No preamble, no recap, no closers. The repo's own summary of the problem is that assistants bury the answer, and every one of the ten rules is a direct countermeasure to a specific habit models picked up from RLHF.
It installs as a plugin marketplace entry: claude plugin marketplace add ayghri/i-have-adhd, or the codex equivalent with --ref main. MIT, 34 commits, fork it and edit the skill definition if you want different rules. That is the whole surface area.
The name is the interesting part and so is the star count. Something that framed itself as an accessibility accommodation for ADHD picked up thousands of stars in a day, which means most of the people installing it do not have ADHD. They just want the model to stop writing four sentences of throat-clearing before the command. The politeness layer that makes a model pleasant in a chat window is pure cost in a terminal where you are reading output between builds.
Which points at what the skill format is actually being used for. Everyone assumed skills would be capability packs, teaching an agent a new domain or a new tool. In practice a large share of what people write and share are behavior patches, changing not what the agent can do but how it acts. Personality is a config file now, and this repo is the cleanest evidence of that yet. github.com/ayghri/i-have-adhd
← Back to all articles
The rules: 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 in minutes, not "a bit". Make wins visible. Report errors matter-of-factly. Cap lists at five items. No preamble, no recap, no closers. The repo's own summary of the problem is that assistants bury the answer, and every one of the ten rules is a direct countermeasure to a specific habit models picked up from RLHF.
It installs as a plugin marketplace entry: claude plugin marketplace add ayghri/i-have-adhd, or the codex equivalent with --ref main. MIT, 34 commits, fork it and edit the skill definition if you want different rules. That is the whole surface area.
The name is the interesting part and so is the star count. Something that framed itself as an accessibility accommodation for ADHD picked up thousands of stars in a day, which means most of the people installing it do not have ADHD. They just want the model to stop writing four sentences of throat-clearing before the command. The politeness layer that makes a model pleasant in a chat window is pure cost in a terminal where you are reading output between builds.
Which points at what the skill format is actually being used for. Everyone assumed skills would be capability packs, teaching an agent a new domain or a new tool. In practice a large share of what people write and share are behavior patches, changing not what the agent can do but how it acts. Personality is a config file now, and this repo is the cleanest evidence of that yet. github.com/ayghri/i-have-adhd
Comments