Part 2

Requirements, Specs, And Prompts

The four-stage pipeline from idea to working software: requirements, specs, structured prompts, agent execution. Most of the people who can't ship are stuck because they're trying to do all four stages at once in their head; this Part separates them.

Start with section 2.1 →
  1. 2.1

    Requirements: What You Actually Want

    Functional, non-functional, constraints. The test for whether a requirement is good enough: can someone build this without asking a follow-up question?

  2. 2.2

    Specifications: What The Machine Actually Needs

    The seven things you add to a requirement to make it buildable: inputs, outputs, rules, edge cases, errors, data, permissions.

  3. 2.3

    Structured Markdown: The Format Coding Agents Read Best

    The format discipline: headers, code fences, in-scope and out-of-scope sections, worked examples. The structure matters more than the prose.

  4. 2.4

    Context Engineering (the Missing Discipline)

    DEEP-DIVE

    [Deep-dive] The principle behind the playbook in 2.6: managing what the agent can see, when, and in what order. The through-line of Part 2.

  5. 2.5

    Prompting Based On Data Shape

    Most bad prompts come from people who don't know what's in their own data. Tables, columns, relationships, edge cases, dirty rows.

  6. 2.6

    The Claude Code Playbook

    The applied playbook for working with a coding agent: setup, the four-phase loop, prompt patterns, session management, scaling, and the failure modes you'll actually hit.