spec

spec is a guided, spec-driven design skill invoked as /spec. It helps a user produce a useful specification for a large feature before any code is written, treating the spec as the contract that drives later execution rather than decorative documentation. The problem it solves is vague requirements that force code to improvise: the skill is deliberately slow during the definition phase and fast during the writing phase, so ambiguities are surfaced and resolved up front.

The workflow runs in four phases. It first gathers project context by reading a memory file (CLAUDE.md, AGENTS.md, GEMINI.md, or README.md) and reviewing the two most recent specs in specs/ to match numbering, language, and conventions. It then clarifies through blocks of three to five concrete questions covering scope, data, integration, persistence, UX and states, risks, and already-closed decisions — using a native multiple-choice tool when available. Finally it builds the spec section by section following a bundled template.md that defines the header, scope (in and out), data model, implementation plan, boolean acceptance criteria, decisions taken and discarded, and risks. Its allowed tools are limited to Read, Glob, Grep, Write, AskUserQuestion, and safe read-only Bash calls (ls, cat, date).

Target users are developers, tech leads, and product-minded engineers starting a large feature who want a disciplined planning step. Use cases include scoping a new feature, capturing deferred work explicitly so it is not slipped in later, and recording the reasoning behind decisions for future maintainers. It writes only into the specs/ folder and does not generate application code.