Better instructions for AI coding agents

Grill my feature idea

Expose unclear requirements and hidden decisions before implementation begins.

Copy-ready prompt

Turn an idea into a buildable specification

Convert a settled product conversation into a concise implementation contract.

Copy-ready prompt

Map an unfamiliar codebase

Build a useful model of modules, data flow, boundaries, and domain language before changing code.

Copy-ready prompt

Find architecture opportunities worth taking

Identify improvements that reduce future change cost instead of merely rearranging files.

Copy-ready prompt

Implement one vertical slice

Deliver a small end-to-end behavior with tests and repository-aligned structure.

Copy-ready prompt

Diagnose before fixing

Reproduce, narrow, and prove a root cause before touching production code.

Copy-ready prompt

Curated skills — from GitHub

Grill Me

Interviews you until the important branches of a plan or decision are resolved.

mattpocock/skills

Grill With Docs

Sharpens a plan while building durable domain language and decision records in the repository.

mattpocock/skills

Improve Codebase Architecture

Finds high-leverage opportunities to create deeper modules and reduce future change cost.

mattpocock/skills

Test-Driven Development

Runs a focused red–green–refactor loop one vertical behavior at a time.

mattpocock/skills

Diagnosing Bugs

Uses reproduction, minimization, hypotheses, and instrumentation before implementing a fix.

mattpocock/skills

Impeccable

A cohesive design language for AI coding agents with commands for shaping, critiquing, polishing, adapting, and hardening interfaces.

pbakaus/impeccable

Practical workflows — step by step

From rough idea to buildable specification

Resolve the important product decisions, cut excess scope, and leave a specification that implementation can follow.

4 steps · 30–60 min

Understand an unfamiliar codebase

Move from repository noise to a working model of the domain, execution path, and modules relevant to your change.

4 steps · 20–45 min

Rescue a codebase becoming a ball of mud

Find the repeated change pain, create a deeper seam, and refactor without turning the effort into a rewrite.

4 steps · 45–90 min

Diagnose and fix a difficult bug

Replace speculative patches with reproduction, hypothesis testing, a minimal fix, and durable regression protection.

4 steps · 30–120 min

Build a high-craft interface from a reference

Extract hierarchy and atmosphere from visual references while keeping the result specific to the current product.

4 steps · 2–6 hours

Curated collections — start here

Real engineering with AI

A disciplined core for alignment, architecture, implementation, tests, diagnosis, and review.

11 resources

High-craft frontend

Product thinking, visual hierarchy, responsive behavior, and restrained finishing for AI-built interfaces.

11 resources

Architecture and codebase health

Shared language, deep modules, focused refactoring, and architecture improvements tied to real change costs.

12 resources

Debugging difficult problems

Reproduction, minimization, instrumentation, feedback, and proof for bugs that resist quick fixes.

8 resources