Descriptions:
Nate Herk demonstrates a Claude Code workflow technique called “Grill Me” — a skill that systematically extracts tacit knowledge from a developer’s head into structured, reusable context files that AI agents can reference across future sessions. Originally designed by Matt PCO as a concise interview-style prompt, Herk’s extended version adds automatic checkpointing: after each question-and-answer exchange, the skill writes a markdown file to a `brainstorms/` folder at the project root, capturing the algorithm, key decisions, and a full Q&A log.
The core problem the technique addresses is the gap between informal brain dumps and the level of specificity an AI system needs to operate consistently. Herk argues that the difference between an 80% and a 95% successful skill comes down to context quality — and that Grill Me can push a new skill to roughly 90% accuracy on the first serious build by having Claude Opus ask targeted, branching questions rather than accepting surface-level answers. The video includes live examples of completed brainstorm files, including one where Claude proactively identified gaps between a session’s findings and an existing packaging skill, then offered to update both.
For teams using Claude Code to automate development or client workflows, the technique offers a practical framework for knowledge transfer: rather than iterating through dozens of failed generations, the investment is made upfront in context extraction. Herk applies it at the project level and also to capture holistic business logic, treating it as an “axe-sharpening” step before building.
📺 Source: Nate Herk | AI Automation · Published June 04, 2026
🏷️ Format: Workflow Case Study
