i-have-adhd Skill Tops GitHub Trending With 3,854 Stars

What Happened

The ayghri/i-have-adhd repository led GitHub trending with 3,854 stars gained in a single day, making it the platform's fastest-rising project over that window. The repo ships an agent skill that constrains coding agents to emit short, answer-first responses rather than verbose reasoning dumps. It is distributed as a skill artifact rather than a model or framework, targeting the output-formatting layer of agent operation.

Why It Matters

Verbose agent output is a tax paid at every step of a coding loop: reviewers skim, downstream tools truncate, and context windows fill with restated reasoning rather than the token that resolves the task. A formatting skill addresses this without touching model weights, retraining, or inference infrastructure, which makes it deployable in minutes across an existing fleet. The rapid star velocity suggests the pain is widely distributed rather than concentrated in one team or stack. For operators running many concurrent agents, output discipline converts directly into reduced token spend and faster human-in-the-loop verification. The strategic read is that behavioral control is migrating from prompt engineering and fine-tuning toward packaged, versioned skill artifacts that can be swapped per task.

Technical Details

The skill is a prompt-layer constraint applied to the agent's system or skill configuration — it reshapes response structure (lead with the answer, defer or omit exposition) without modifying the underlying model. It targets the general class of coding agents that accept skill or instruction files, so integration is a configuration change rather than an SDK dependency. There is no benchmark suite or token-savings figure published in the summary, and no versioning or provider-compatibility matrix is stated; treat the 3,854-star figure as an adoption signal, not a quality measurement. Formatting enforcement of this kind is probabilistic — models can drift from the constraint under long contexts or adversarial inputs, so operators should expect occasional regressions in output shape. The skill's scope is output shaping only; it does not address tool-call verbosity, multi-step planning traces, or retrieval behavior.

Operational Impact

Teams can add a formatting constraint to existing agents and see changes within a single deploy cycle, with no GPU cost and no eval retraining. The immediate workflow effect is shorter PR-review surfaces and lower per-interaction token counts, which compounds across high-volume agent runs. It also reduces the need for post-hoc output parsers and truncation logic that operators currently maintain to make agent responses usable. The tradeoff is a loss of intermediate reasoning visibility — if the model suppresses its steps, debugging failures becomes harder, so operators should pair this with separate trace logging rather than relying on the response body. Expect skill-marketplace dynamics to follow: formatting skills become commodities, and differentiation shifts to task-specific skills with measurable correctness gains.