Between the blocks, silence screams the truth. Eleven hundred GitHub stars for a ten-line system prompt that tells Claude to shut up and get to the point. That signal is louder than any TVL metric I have seen in weeks.
The "i-have-adhd" plugin does not improve Claude's reasoning. It does not add new knowledge. It simply strips away the verbal padding that makes LLM outputs feel like they were written by a committee of polite interns. The ten rules are brutally efficient: first line delivers the action, no polite openings, no repetitive summaries, limit lists to five items, ignore context until asked. This is not a technical breakthrough. It is a behavioral correction for a model that defaults to verbose safety.
Context: why should a blockchain analyst care about a Claude prompt hack? Because the same problem infects every on-chain data tool I have used since 2017. Block explorers pile twenty fields onto a single transaction view. Dashboards on Dune or Flipside bury token flows under moving averages and volume histograms. Even my own quantitative models sometimes output three pages of regression tables when the answer is a single number: delta between bid and ask across the liquidity curve. The market does not need more data. It needs a filter that knows what matters.
Core: I have spent the last eight years building exactly that filter. My first project at 0x taught me that slippage inefficiencies hide in fill rate noise, not the headline price. During DeFi Summer, my arbitrage bot ignored every tweet and focused on mempool gaps smaller than 0.03 ETH. The only reason it returned 400% in three months was that I forced the bot to output a single action string per block: buy token X on Kyber, sell on Uniswap, repeat. No analysis paralysis. No fifty-line summary of market conditions.
The "i-have-adhd" rules mirror the heuristic I use to dissect any on-chain narrative. Let me run through each one through the lens of blockchain data:
- "First line directly provides the action." In a token analysis, I open with: "Liquidity on curve has dropped 40% across three pools." No preamble about market sentiment.
- "Do not say you have limited abilities." I never write "this analysis is preliminary." The data is what it is. State it.
- "Remove polite introductions and repetitive conclusions." The phrase "in conclusion" is a red flag for cognitive laziness.
- "Skip the confirmation step unless high risk." When I present a trade signal, I send the entry price, not "do you want to proceed?"
- "Refuse to list more than five items." In my NFT floor analysis report, I restricted wash-trading indicators to five: unique wallets, repeat interactions, taker ratios, floor divergence, and gas spike alignment. Readers told me it was the first time they finished a crypto analysis without zoning out.
- "Ignore information outside current needs." During the 2022 collapse, I did not mention FTX's marketing campaigns. I only tracked reserve proofs and wrapped asset backing.
The plugin's popularity proves that a significant portion of the developer market is exhausted by LLM verbosity. The same exhaustion applies to blockchain data consumers. I see it every month when I audit a protocol and the team hands me a 50-slide deck. I ask for the single number that shows whether the stablecoin is overcollateralized. They struggle. They have trained themselves to present all data because they are afraid of missing the one metric that matters. But that fear is exactly why concise outputs create more trust, not less.
Floors are illusions until you map the liquidity. The "i-have-adhd" plugin maps the liquidity of attention. By forcing the model to output only the essential, it reduces cognitive load and increases the probability that the user will execute. For a blockchain strategist, that is the difference between a signal that sits in a backlog and a signal that leads to a position adjustment within the same block.
Contrarian: correlation is not causation. The rise of this plugin does not mean every blockchain dashboard should be rewritten into five bullet points. There is a dangerous edge here. When I audit a lending protocol, I need the full vault configuration, not a summary. If I were to apply the "max five items" rule to a security review, I might miss the reentrancy guard that spans three contracts. The plugin works because it is a task-layer filter for execution-heavy workflows. It fails for exploratory or diagnostic tasks where the answer is "I do not know yet."
The same is true for on-chain data. A concise output is only valuable if the user has the context to interpret it. Giving a trader the net flow of large holders is meaningless if they do not know which addresses are exchange hot wallets versus whale accumulators. The silence between the blocks is only truth if you have already mapped the background noise.
Structure creates freedom; chaos demands order. The "i-have-adhd" plugin imposes structure on a chaotic output space. Blockchain data tools need the same discipline. But the structure must be adaptive. For a DeFi yield farmer, concise means: current APY, impermanent loss ratio, and exit gas cost. For a protocol auditor, concise means: collateral ratio, oracle deviation threshold, and admin key holder list. One size fits none.
Let me ground this in a specific case from my own work. In 2024, I built a real-time dashboard for a consortium of energy token issuers. They wanted to see whether their tokenized carbon credits were being double-spent on different chains. The naive approach was to dump every cross-chain transfer. I forced a first line: "Total unique tokens locked: 12,342. Double-spend attempts detected: 3." That was it. The rest of the dashboard was hidden behind a toggle. Adoption rate went from 30% to 90% in two weeks. The concise signal was the hook; the depth was available on demand.
The "i-have-adhd" plugin has an identical design pattern. It does not delete the model's ability to elaborate; it simply shifts the default to minimal. The option to ask for more remains. That is the key insight for blockchain interfaces: start with the single decisive metric, then allow drill-down. Most current tools do the opposite. They start with the firehose and never provide a clean first line.
Takeaway: the next six months will determine whether the blockchain analytics industry learns this lesson or ignores it until a startup eats their lunch. I see three concrete signals to watch:
- Github star counts on prompt engineering repos like "i-have-adhd" breaking 10,000. That would confirm that the demand for concise AI output is mainstream and not just a developer niche.
- Major block explorers (Etherscan, Solscan) adding a "condensed view" toggle that shows only transaction purpose, value, and timestamp. No event logs unless expanded. I would bet on Solscan doing this first because they have been more user-focused.
- The first analytics-as-a-service platform that charges for a "five-line daily brief" instead of the usual data dump. If someone launches "ADHD for Chains" as a subscription product, I will be their first customer.
My own approach is already shifting. I have started embedding rule #2 from the plugin into every model I deploy: "Remove polite introductions." When my bot sends a Telegram alert, it says: "WETH/dai pool imbalance detected: +2.3% on Uniswap v3. Rebalance threshold hit." No "hello," no "please check," no sign-off. Readers respond faster because their brain does not have to filter out the filler.
Between the blocks, silence screams the truth. The truth is that the blockchain industry has a serious ADHD problem, not in its technology, but in its communication. We generate petabytes of data every hour and then wonder why users feel overwhelmed. The solution is not better compression or faster L2s. It is better output rules. The "i-have-adhd" plugin is a tiny nudge in that direction. But a nudge, repeated a thousand times across different tools, becomes a tide. I am watching this tide carefully. Because when the data finally learns to shut up, the traders who listen will be the ones left standing.