Skip to content

Every Claude Code Hack I Know (March 2026)

Author/Source: @mvanhorn (Matt Van Horn), X/Twitter thread, March 22, 2026

Key Ideas

  • The core workflow is "plan first, always": every idea immediately becomes a plan.md file via /ce:plan, flipping the traditional 80/20 coding-to-planning ratio
  • The Compound Engineering plugin enables a plan-then-build loop where /ce:plan launches parallel research agents and /ce:work implements the plan with test verification
  • Voice input (via Monologue or WhisperFlow) is a critical productivity multiplier because Claude Code is smart enough to interpret imperfect transcription
  • Running 4-6 parallel Claude Code sessions simultaneously is the author's default working mode, requiring bypass permissions and audio notifications to manage
  • Three essential settings: "dangerously skip permissions" for autonomous operation, sound hooks for completion alerts, and Zed autosave at 500ms for real-time co-editing
  • /last30days skill searches Reddit, X, YouTube, HN, and more in parallel to ground plans in current community knowledge rather than stale training data
  • Meeting transcripts (via Granola) can be turned into structured product proposals by cross-referencing against the codebase and prior strategy documents

Summary

This article is a practitioner's guide to power-using Claude Code, written as an X/Twitter thread by a software developer and open source contributor. The central thesis is that AI-assisted work should be plan-driven: every idea, bug, or feature request immediately becomes a structured plan.md file before any implementation begins. The Compound Engineering plugin operationalizes this by launching parallel research agents that analyze the codebase, search past solutions, and produce a grounded plan with acceptance criteria. Implementation then becomes mechanical execution of the plan via /ce:work.

The article describes an intensive workflow: 4-6 parallel Claude Code sessions running simultaneously in separate terminal windows, with voice dictation replacing typing, bypass permissions enabling autonomous operation, and audio notifications signaling task completion. The author runs /last30days research queries before planning to ground decisions in current community knowledge. Meeting transcripts from Granola are processed into structured proposals by cross-referencing against codebases and prior strategy documents. The piece concludes with a Disney World trip-planning example that demonstrates the full workflow (research, plan, build, deploy, automate reminders) applied to a non-coding task, emphasizing that the approach is domain-agnostic.

Relevance to Economics Research

While written from a software development perspective, several techniques transfer directly to economics research. The plan-first discipline maps onto research design documents and pre-analysis plans. The /last30days research tool could be used to survey current discussions around methodological debates or policy topics before starting a project. Voice-driven workflows reduce friction for researchers who think faster than they type. The meeting-to-proposal pipeline (Granola transcript to structured document) is relevant for converting seminar discussions or co-author meetings into actionable research plans. The parallel-session approach suits researchers juggling multiple papers simultaneously.