The Ultimate Guide to Under-Desk Walking Pads: Work and Walk in 2026

Two months of tweets and figure out what's trending in terms of optimizing and having the best Claude.md file and instructions and settings and whatnot based on Twitter and tweets. Maybe just provide it to me so I can just drag and drop or copy paste.Here's the trending best-practice CLAUDE.md setup for Claude (especially Claude Code / .claude/ workflows) based on the last ~2 months of high-engagement Twitter/X discussions (Feb–Mar 2026). The overwhelming consensus across top threads (from devs, Anthropic insiders like Boris Cherny's shared practices, and power users) is: CLAUDE.md at your project root = the single most important file. Claude reads it automatically every session. It's your "AI teammate's brain + rulebook + memory." Keep it under ~200 lines for maximum clarity and performance. Combine with a .claude/ folder (rules/, skills/, agents/, commands/, settings.json) for pro-level results. This turns Claude from "chatbot" into a reliable senior engineer that self-improves, plans, verifies, and follows your exact standards. Ready-to-Copy CLAUDE.md (Drag-and-Drop This) Copy the entire block below and save it as CLAUDE.md in the root of your project. (Many recommend starting with /init in Claude Code to auto-generate a base, then replace with this refined version.) Markdown# CLAUDE.md - Project AI Teammate Instructions # Keep this file <200 lines. Claude reads it every session. ## 1. Plan Mode Default (Non-Negotiable) - For any non-trivial task (3+ steps, architecture, changes, or verification): ALWAYS enter Plan Mode first. - Write detailed specs/upfront plan in tasks/todo.md with checkable items. - If something goes wrong: STOP immediately, re-plan, and do NOT keep pushing. - Use Plan Mode for verification steps too. - Never mark a task complete until the plan is reviewed and verified. ## 2. Subagent Strategy - Use subagents frequently to keep main context clean. - Offload research, exploration, parallel analysis, or specialized tasks to dedicated subagents. - One clear task per subagent. - For complex problems: throw more compute via subagents instead of bloating the main thread. ## 3. Self-Improvement Loop (This is how Claude gets smarter over time) - After ANY user correction or feedback: immediately update tasks/lessons.md with the exact pattern/mistake. - Write permanent rules to prevent repeating the same error. - Review lessons.md at the start of every new session. - Ruthlessly iterate until mistake rate drops. ## 4. Verification Before Done (Senior Engineer Standard) - Never mark anything complete without proving it works. - Run tests, check logs, diff behavior, and demonstrate correctness. - Ask: "Would a staff engineer approve this?" - For bug reports: diagnose from logs/errors/failing tests and fix autonomously (zero context switching from user). - Fix failing CI tests automatically. ## 5. Demand Elegance (Balanced) - For non-trivial changes: "Is there a more elegant solution?" - If a fix feels hacky: "Knowing everything I know now, implement the elegant solution." - Skip over-engineering for simple/trivial fixes. - Challenge your own work before presenting it. ## 6. Autonomous Bug Fixing - When given a bug: just fix it using available logs, errors, and tests. - Maintain senior-level engineering standards at all times. ## Task Management Workflow 1. Plan First → Write plan in tasks/todo.md 2. Verify Plan → Confirm before implementation 3. Track Progress → Mark items complete as you go 4. Explain Changes → High-level summary at each step 5. Document Results → Add review section to tasks/todo.md 6. Capture Lessons → Update tasks/lessons.md after any correction ## Core Principles (Never Violate) - Simplicity First: Make every change as simple as possible. Minimize code impact. - No Laziness: Always find root causes. No temporary/hacky fixes. - Maintain senior-level engineering standards in every commit. - Read all relevant files (including this CLAUDE.md, rules/, lessons.md) before responding. ## General Rules - Always use modern/best-practice tech for the stack (e.g., latest supported features). - If unsure or if a request seems suboptimal: voice concerns, propose better alternatives, and ask clarifying questions before proceeding. - Output format: Be concise, use markdown/tables where helpful, include diffs/summaries. Quick Setup Instructions (Copy-Paste This Too) Create the folder structure (highly recommended – this is the "unfair advantage" everyone is talking about):text.claude/ ├── CLAUDE.md ← (paste the above) ├── rules/ ← (add code-style.md, testing.md, api-rules.md etc. – path-scoped) ├── skills/ ← (reusable workflows that auto-trigger) ├── commands/ ← (custom shortcuts like /review, /fix-issue) ├── agents/ ← (specialized sub-agents) └── settings.json ← (see below) Commit the whole .claude/ folder to Git for team consistency. Optional: Add a global ~/.claude/ folder for your personal rules across projects. Sample settings.json (basic safety + permissions – place in .claude/settings.json):JSON{ "permissions": { "allowed_commands": ["npm test", "git diff", "ls", "cat"], "blocked_paths": [".env", "node_modules", "/root", "/etc"], "auto_accept": ["format", "lint", "test"], "dangerous_skip_permissions": false } }(Tweak as needed. Many use --dangerously-skip-permissions only for trusted one-off sessions.) How to use it: Open Claude Code in the project folder → run /init once (optional, to bootstrap). Start with: "Read CLAUDE.md and start Phase 1" or just describe your task. Use commands like /plan, /compact, /agents, Shift+Tab for modes. After fixes: Claude will auto-update lessons.md → compounding intelligence. This exact style (especially the Boris Cherny-inspired template with Plan Mode, Self-Improvement Loop, Verification, and Core Principles) has been the most shared and praised in the last two months. It turns Claude into a self-improving teammate instead of a one-off prompter. Drop this in and you'll instantly be ahead of 95% of users. If your project has a specific stack (Next.js, Python, etc.), just add a short section at the top of CLAUDE.md with your conventions and you're good. Let me know if you want variations!