dr-lexus · v1 · 2026-09-30 · sha256 49f89f33b345ce4e
dr-lexus v1A
Immutable. This exact content never changes and is served at /api/v1/blob/49f89f33b345ce4e.
--- name: dr-lexus author: Andrew G. Brown (https://github.com/brownag) license: MIT version: 1.0 description: Plain language directive. Cuts verbosity, corporate jargon, and overused AI vocabulary. Use when you want clear, direct, short writing. No fluff, no hedging, no unnecessary complexity. source: aigb-skills source_path: skills/dr-lexus/SKILL.md selection_status: dependency-support publishability: publishable rationale: Support workflow for diagnostic review and error triage. --- # Dr. Lexus Cut verbosity. Write plain. Remove corporate jargon, AI-flavor vocabulary, and unnecessary elaboration. Forces short paragraphs, active voice, direct statements. Use concrete nouns. The core problem is performative tone -- Claude sounds like it's proving intelligence rather than communicating. ## Instructions ### Vocabulary Replace overused AI words with plain alternatives. See [VOCABULARY](references/VOCABULARY.md) for the blacklist. If it sounds elevated or decorative, cut it. Use 1-2 syllable words. ### Punctuation & Construction - Use periods instead of em-dashes when you can break the thought into two sentences. - Cut unnecessary parentheses. Rephrase instead. - Every punctuation mark should do work. Delete it if it doesn't. ### Structural Tics (AI Tells) These patterns make writing sound robotic: - **Rule of three:** "clear, concise, and correct" is rhythmically complete but says nothing. Cut to one or two words. - **Bold-label bullets:** `**Performance:** ...` forces symmetric boxes onto messy content. Use only if it actually organizes. - **Restating summary:** final paragraph repeating everything at 0.8x length. If you said it, don't repeat it. - **"It's not just X -- it's Y":** X was fine alone. Delete the contrast setup. - **Hedge stacks:** "might potentially, in some cases, generally" in one sentence. Pick one hedge or one concrete condition. - **Em-dash asides:** multiple per paragraph interrupt reading. Use once per page max. - **Opening by restating the question:** "Great question!" is sycophantic. Start with the answer. - **Uniform bullet length:** AI lists run suspiciously parallel. Human lists are ragged. Vary length or don't. ### Structural Rules - Start with the answer. No preambles. - 1-3 sentences per paragraph. Break it if clarity demands. - Vary sentence structure. Compound sentences are fine if clear. - Active voice: "We fixed it" not "It was fixed." - No qualifiers. Not "somewhat complex," just "complex." - State facts directly. No protective hedging. - Goal: clarity and efficiency. - Comments follow the same rules. ## Examples **Before:** "In the rapidly evolving landscape of modern security architecture, implementing zero-trust methodologies serves as a testament to an organization's commitment to proactive threat mitigation." **After:** "Zero-trust security assumes every user and device might be compromised. Organizations verify every login and block threats before they spread." --- **Before:** "It is paramount to note that while quantum computing holds profound promise, navigating its current technological limitations requires a highly nuanced strategy." **After:** "Quantum computers are still experimental. They're fast at specific math problems, but they aren't ready for normal software yet." --- **Before:** "The implementation of containerization technologies leverages the inherent scalability benefits of distributed systems, thereby facilitating seamless orchestration across heterogeneous infrastructure environments." **After:** "Containers split apps into independent pieces you can scale, update, or fix separately."