Skip to main content
  1. Projects/

specy

specy is a Claude Code plugin that runs /spec — a guided interview that walks from a fuzzy idea, or an existing codebase, to a single authoritative SPEC.md a coding agent can build from. It doesn’t just ask questions; it conducts a sequence of curated methods (success definition, competitive analysis, prior-art research, constitution, codebase mapping, gap analysis, persona voicing) through one interrogation, then synthesizes the answers into one document.

How it’s built
#

specy is self-contained — it ships its own slimmed-down versions of each method skill so it works standalone with nothing else installed. It runs amplify-first: if the full versions are installed from the public ai-skills library, specy prefers those automatically and falls back to its bundled versions otherwise. Heavier standalone tools — abe for multi-model debate, superpowers:writing-plans for the final plan hand-off — are used if present, but nothing is required.

The 12 phases
#

Frame → Why ×5 → Users & Jobs → Success → Compete & Differentiate → Vision & Non-negotiables → Features & Scope → Design (how it works) → Stack → Architecture → Persona Review → Synthesize. Each phase runs a bundled method skill; optional tools like abe sharpen a phase further when installed. State lives in SPEC.md’s own YAML frontmatter, so the interview can be quit and resumed at any point — the filesystem is the database.

Usage
#

/spec "a tool that turns ideas into specs"   # idea mode
/spec ./my-project                            # codebase mode — maps current state first
/spec status                                  # what's answered / stubbed
/spec resume                                  # pick up where you left off
/spec jump stack                              # jump straight to a phase
/spec "an idea" --brutal                      # AAT-style critique, plus abe debates if installed

Default temperament is assertive — it pushes back on vague answers rather than accepting them. Dial it with --gentle or --brutal.

What it’s for
#

  • Turning a fuzzy idea into a build-ready spec — one document a coding agent can implement directly from, instead of a scattered thread of decisions.
  • Mapping an existing codebase before extending it — codebase mode runs a reverse-engineering pass first, so the interview starts from what’s actually there.
  • Stress-testing a plan before committing to it--brutal mode brings in adversarial critique and multi-model debate on the weakest points.

Links#