Ed Harrod

ALT Before/after diff of ~/.claude/CLAUDE.md from Anthropic's Opus 5.5 guide. Before: think carefully, drip-feed the steps, stopping rules in the prompt, skim the summary, describe the look you want. After: nothing (it sets its own depth), whole task with "done means" spelled out, stopping rules in CLAUDE.md, answer its questions first, name the patterns you don't want.

ALT ByteByteGo infographic, Top 9 places to use Jev: model routing, guardrails, tool-call gating, inbox triage, reranking, LLM evals, bulk labelling, real-time control, confidence gates.

ALT Decision Index 0.3.1 bar chart ranking 143 open reproductions of Jev's decision model by full score.

ALT Pixel-art banner reading 'Claude Superpowers plugin', with an arrow pointing to an orange pixel-art Claude mascot.

ALT Two-panel meme. Top, 'Prompting Claude Code directly': a figure steps on a rake and it smacks them in the face. Bottom, 'Using spec-kit': a man rides a rake down a flight of stairs like a skateboard, pulling off tricks, before it smacks him in the face at the bottom.

ALT Flowchart of our AI workflow. A new work item is sorted by complexity. Small: Viktor or Claude completes it directly in Slack. Medium: Viktor or Claude creates a GitHub issue and a developer picks it up with Claude Code. Large: a spec.md is written and reviewed by the team, then the developer runs /speckit.plan, /speckit.tasks and /speckit.implement. Every path ends in an MR that CodeRabbit reviews, with an agent committing and resolving comments until there is no more feedback.