Recipes

Each recipe starts from a problem you keep losing time to and shows the smallest skill, or combination of skills, that deals with it. The aim is to settle the first decisions early, so less work has to be redone later. Every skill also works alone, and Choose a skill maps single situations to single skills.

• Plan a change before your agent starts editing: decisions get made halfway through editing and the work is redone.
• Fix the bug without patching the symptom: a fix lands where the symptom shows and the bug comes back.
• Settle the structure before agents add parallel machinery: agents add a second way to do something the code already does.
• Work through a large ticket without losing the thread: work spread over sessions keeps being rediscovered or left half-done.
• Trust a change before you merge it: reviews skim the diff and fixes quietly reopen decisions.
• Find the next step when work spans several skills: the next skill is unclear and the agent drifts.
• Get a stuck decision moving again: a decision stalls or rests on an unchecked assumption.
• Write documentation that matches its sources: a document reads well but cannot be followed.
• Learn a subject and practise for its exam: learning and examination practice drift apart.

Each skill's own page, in the skill index, says what it does and where it stops.

The worked scenarios

Each recipe ends with a worked scenario. The coding recipes all follow one made-up project, FreshCart, an online grocery service with orders, payments, customer messages, stock, and search, so you can watch one codebase move from a bug report through planning and review to a pull request. The learning recipe follows a student instead.

The scenarios are illustrative. The prompts, findings, and file excerpts were written to show what each skill does and leaves behind; they were not recorded from a real run. Excerpts are shortened. Anatomy of a sealed blueprint walks through one complete FreshCart plan.