Hook
A simple GitHub repository with 1,100 stars. Ten rules. No new model weights, no cryptographic proofs, no tokenomics. Yet the “i-have-adhd” Claude Code Skill has become a quiet inflection point in how developers interact with large language models. It is not the code that matters—it is the revelation that the last mile of AI deployment is not intelligence, but formatting. For a blockchain protocol developer, this pattern is hauntingly familiar. The same obsession with output efficiency, the same blindness to systemic fragility, and the same gap between user desire and architectural integrity. This article dissects the plugin through a seven-dimensional lens borrowed from my own protocol audit framework, and then extracts the hidden signals for the blockchain world.
Context
The “i-have-adhd” skill is a set of ten system prompt instructions designed to make Claude’s responses short, direct, and actionable. Rules include “first line must be the action,” “no polite openings,” “max five bullet points,” and “skip summaries.” It does not enhance Claude’s reasoning; it only constrains output format. It is a prompt engineering template, not a model fine-tune. The repository is open source, community-maintained, and licensed permissively. Its success reflects a widespread frustration with verbose, over-polite AI replies, especially in developer workflows where every extra word is noise.
But beneath the surface, this is not just a tool for AI. It is a mirror for the very same trade-offs that haunt DeFi composability, L2 data compression, and stablecoin peg mechanisms. The blockchain world has its own “i-have-adhd” moment every time a protocol sacrifices redundancy for speed, or audits for gas optimization. The following seven-dimensional analysis applies my own post-mortem methodology—honed from the Solidity audit of 2017 to the collapse of Terra—to this apparently trivial AI plugin. The goal is to show that the same reasoning errors appear across domains, and that understanding them in one field inoculates against them in another.
Core: Seven-Dimensional Technical Dissection
Dimension 1: Technical Route
The plugin is a pure inference-time output control mechanism. No model architecture change, no training data shift. It is analogous to a block-level gas limit adjustment in a blockchain—it changes how data is packaged, not what data is validated. The ten rules are essentially a hardcoded “reply template” that Claude must follow. The hidden insight is that the rules exploit the known bias of transformers toward recency and specificity. By placing action first, the model is forced to generate the most salient token immediately. This is clever, but it introduces a brittle assumption: that every task benefits from brevity. In protocol terms, it is like setting a fixed block gas limit without considering the size of each transaction. Some tasks will overflow and truncate poorly.
Dimension 2: Commercialization
Zero direct revenue. No API keys sold. No premium tier. But the indirect commercial value is enormous. The plugin increases user stickiness to Claude’s platform. It also sends a clear signal to Anthropic about a gap in their default product. Anthropic can—and likely will—incorporate these rules into their official system prompt, killing the plugin’s reason for existence overnight. This is the classic “platform risk” that every third-party DeFi aggregator faces. The blockchain parallel is a smart contract that wraps an external protocol’s functions. If the protocol changes its interface, the wrapper becomes obsolete. The commercial moat is zero, as the analysis correctly notes with confidence C.
Dimension 3: Industry Impact
This plugin will accelerate the standardization of prompt engineering practices. Developers will adopt similar rule sets across different AI tools. The first-order effect is improved productivity for task-oriented coding. The second-order effect is a potential homogenization of AI output—every assistant starts sounding like a terse command line. The third-order effect, which I find most disturbing, is the loss of nuance in safety-critical contexts. In blockchain terms, this is the equivalent of standardizing a gas-efficient but insecure reentrancy guard pattern. Standardization improves adoption but can ossify vulnerabilities. The industry impact is high (confidence A) because the change is incremental and universally adoptable, just like the adoption of OpenZeppelin’s initial Ownable contract.
Dimension 4: Competitive Landscape
This plugin does not compete with other AI models; it complements Claude. Its real competition is other prompt libraries, such as those on GitHub for GPT-4 or Gemini. With only 1,100 stars, it is a niche artifact. Confidence D because the competitive dynamics are trivial. However, for blockchain readers, this is a cautionary tale: a protocol that relies on a non-differentiable UX tweak (like a nicer UI for Uniswap) will never beat the underlying platform. The real competition is against the default experience.
Dimension 5: Ethics & Safety
The analysis flags a low but real risk: concise output may skip safety warnings. For example, a rule that says “skip summaries” could also skip a mandatory disclosure about generated code being insecure. My own audit experience with flash loan aggregators taught me that efficiency optimizations often accidentally optimize away safety checks. The name “i-have-adhd” is also ethically ambiguous—it uses neurodivergence as a branding hook, which could trivialize ADHD. Confidence C.
Dimension 6: Investment & Valuation
Not applicable. The plugin has zero investment, zero revenue, zero token. Confidence E.
Dimension 7: Infrastructure
The plugin depends entirely on Claude’s API and the user’s terminal. No new infrastructure. Confidence E.
Contrarian: The Blind Spot of Output Optimization
The mainstream narrative celebrates this plugin as a win for user experience. I see the opposite: it is a symptom of a deeper problem where AI assistants are trained to be sycophantically polite and verbose, and the only fix is a hack. The plugin is not a solution; it is a bandage. The real fix would be to train models that natively understand when to be brief and when to elaborate, based on the user’s intent and the complexity of the query. In blockchain, the equivalent is blaming high gas fees on user behavior and attacking them with gas-efficient patterns, rather than fixing the underlying state bloat caused by excessive data storage. Fragility is the price of infinite composability—and infinite rule application without adaptation is fragility.
Takeaway: The Signal for Blockchain Protocol Designers
The “i-have-adhd” phenomenon teaches us that in any technical system, the most vocal user demand is often for marginal efficiency gains, not fundamental robustness. For blockchain, this translates into an over-focus on reducing transaction sizes, lowering gas costs, and optimizing VM opcodes, while ignoring systemic risks like centralized sequencer failures, data availability timeouts, and governance capture. The next crisis will not come from a lack of efficiency—it will come from a collapse of resilience hidden beneath the veil of streamlined output. As I wrote in my post-mortem of Terra, the algorithm worked perfectly until it didn’t. The same will apply to any chain that prioritizes “direct action” over “complete audit trail.” The question is not whether your protocol is fast—it is whether it can survive when the rules change.