Hook
We do not build for today. Last week, a GitHub repository named 'i-have-adhd' crossed 1,100 stars. It is not a protocol upgrade, not a new consensus mechanism. It is a set of ten rules that force Claude's output to be concise. As a Core Protocol Developer who has spent years auditing smart contracts, I see something deeper than a productivity hack. This skill reveals a fundamental tension between AI-generated verbosity and the precision required in blockchain infrastructure.
Context
'i-have-adhd' is a Claude Code Skill — a system prompt template that overrides the model's default conversational tone. Its ten rules include: 'First line provides action – no preamble,' 'Remove polite greetings and repetitive summaries,' 'List at most 5 items,' and 'Only retain currently needed information.' It does not modify Claude's reasoning; it only constrains output format. The creator aimed to reduce cognitive load for task-oriented users, particularly developers who want direct answers.
The skill's popularity signals a demand for AI that respects the user's time. But in blockchain development, time is often the enemy of security. The art is the hash; the value is the proof. A brief response can hide a fatal vulnerability.
Core
Let me dissect the rules through the lens of a cryptographic audit. Rule #1: 'First line provides action.' In a typical code review, I want the AI to first state the risk — say, 'Reentrancy vulnerability in function withdraw().' That action line is useful. But rule #2: 'Remove polite greetings and repetitive summaries' might strip away critical disclaimers like 'This is not a full security audit; always test in isolation.' Reentrancy doesn't forgive shortcuts. If the AI omits context, a developer might treat a superficial suggestion as exhaustive.
Rule #3: 'List at most 5 items.' For a complex Solidity contract, five items may not cover all attack vectors. During my 2018 Parity Wallet audit, I found a logic flaw in ownership update that required tracing six state transitions. A five-item limit would have truncated the explanation. The skill's implicit assumption — that brevity equals clarity — is dangerous when the cost of error is millions of dollars locked in a DeFi protocol.
Rule #7: 'Only retain currently needed information.' Blockchain debugging often requires cross-referencing historical states. If the AI discards previous context to stay concise, it loses the thread of a complex reentrancy chain. I have seen tools that prune context and then suggest incorrect fixes because the model forgot the initial storage layout.
The skill's true technical contribution is its recognition that default AI outputs are too verbose for task execution. But its solution is a blunt instrument. As an architect, I prefer adaptive systems: a toggle between 'exhaustive' and 'concise' modes, not a permanent restriction.
Contrarian
Here is the contrarian angle: the 'i-have-adhd' skill is a canary in the coal mine. Its popularity shows that even sophisticated developers find current AI models unhelpful in their default state. But embracing such rules risks training ourselves to accept incomplete answers. In blockchain, where formal verification and exhaustive testing are ethical obligations, we cannot afford to normalize shortcut outputs.
The skill's name — referencing ADHD — is a clever metaphor for the modern information overload. Yet it also stigmatizes the need for thoroughness. A secure smart contract audit is not a quick checklist; it is a deep dive into every execution path. We do not build for today; we build for eternity.
I predict that Anthropic will soon incorporate these rules into Claude's official system prompt, rendering the third-party skill obsolete. But that integration will come with nuance — perhaps an 'urgency' parameter that dynamically adjusts output density based on task risk. Until then, developers should use such tools with caution. Always verify AI-generated code against the white paper. Always demand the full proof.
Takeaway
The 'i-have-adhd' skill is a mirror reflecting our collective impatience. As blockchain infrastructure matures, we must resist the temptation to prioritize speed over substance. The next time an AI offers you a one-line solution, ask yourself: Is the proof complete? Or is it merely concise?