Prompt Engineering

Four documents · verbatim

Prompts that do real work

Not examples written to teach. These are the documents AI agents actually read before doing a job — kept because the thinking in them transfers to anything you would ask a model to do.

Every page has Highlights and the Full original. Nothing is reworded: passages are cut from the source files by line number, so a highlight and its original are the same bytes.

01

The Philosophy

How to think about building software, and about what to hand an AI instead of encoding

Judgement belongs to the agent. Code moves bytes. · Nothing is added between the row and the worker. · Prefer a structural fact to a check.

24 passages

 02

Training Jess

The craft of editing a prompt: which instrument to reach for, and which ones backfire

Removal is the default · The competence pass · When the fix is not words

50 passages

 03

The Codebase Audit

A whole engineering team, briefed as one prompt

Phase 3: Set the standard and the north star · Phase 1: Map the codebase · Phase 2: Analyze and report (no implementation)

9 passages

 04

Further Principles

From four more design plans, recovered from git history

Step 5 — prove the seam with a throwaway · The instruction that matters more than any other · Target the same size — around 85 lines

17 passages