Pinned
Cory House
@housecor
I help dev teams be insanely productive with AI.
Courses: pluralsight.com & dometrain.com
Consulting: reactjsconsulting.com ⚛️
- I think about prompts as a spectrum. Too low level: "Change line 24 to a while loop." Too high level: "Implement auth." Syntactic prompts are too low level. A feature in a sentence is too high level. Wisdom is in the middle.
- Programming language: A human-readable computer instruction format that's typically generated by AI.
- Problem: LLM instructions get bloated, inaccurate, and out of sync over time. They also need to be audited for opportunities to extract skills, subagents, and chances to progressively disclose instructions in separate files. Solution: 1. Create an audit-instructions skill that
- Trend I've noticed: People using instructions when they should consider a skill instead. Example: Putting a bunch of PR rules in AGENTS/CLAUDE md. Instead, create an open-pr skill. Then the instructions are lazy loaded, and can be invoked on demand.
