Skip to content

README.md

Latest commit

 

History

History
119 lines (86 loc) · 4.9 KB

README.md

File metadata and controls

119 lines (86 loc) · 4.9 KB
Raw

cloneify

Turn "I want to clone X" into a single paste-ready prompt that builds it.

cloneify is a Claude Code skill that conducts a four-phase, human-gated pipeline for cloning a piece of software. You stay in the driver's seat — every phase stops for your go-ahead — and the pipeline composes two other skills rather than reinventing them.

The real deliverable is Phase 3's goal-prompt: one self-contained block you paste into a fresh Claude Code session to actually build the clone. Phases 0–2 exist to make that prompt correct — scoped, researched, and adversarially hardened.

The four phases

Phase Name What happens
0 Scope One-question-at-a-time interview: platform, must-have features, hard constraints (the reason to clone), stack, explicit out-of-scope. → scope.md
1 Research Invokes a deep-research skill to understand the target deeply — real feature set, technical approach, existing clones, and the landmines most likely to sink the build. → research.md
2 Plan Invokes grill-me-codex: Claude interviews you to lock every open decision, then OpenAI Codex adversarially reviews the plan in a read-only sandbox until APPROVED. → PLAN.md + PLAN-REVIEW-LOG.md
3 Handoff Emits GOAL-PROMPT.md — a self-contained build prompt (full plan inlined) plus a matching /goal completion condition so Claude Code builds autonomously until it's met.

Gates are hard stops. Each phase produces an artifact, shows it to you, asks the gate question, and waits. You can always say "go back" and revise the prior phase.

Why it works

  • Scope before research — research has a target, and the plan has guardrails.
  • Research before planning — decisions rest on sourced facts, not vibes.
  • Grill + cross-model review before building — the #1 failure is building the wrong thing (fixed by the grill); the #2 is a plan that sounds right but breaks (fixed by a different model, Codex, attacking it).
  • A stand-alone goal-prompt — the build session doesn't need this run's files; the plan and scope are inlined into the prompt.

Prerequisites

cloneify is a conductor — it orchestrates other tools. You need:

  • Claude Code with skills enabled.
  • A deep-research skill available in your environment (Phase 1 invokes it by name). Any deep-research harness that returns a sourced report works; swap the Phase 1 call for your own if you use a different one.
  • The grill-me-codex skill (Phase 2 invokes it by name).
  • The Codex CLI installed and authenticated (codex login) — grill-me-codex uses it for the adversarial plan review.
  • For the optional autonomous build: Claude Code v2.1.139+ with /goal.

Without a deep-research skill and grill-me-codex (+ Codex CLI) present, Phases 1–2 won't run. cloneify does not bundle them.

Install

Drop SKILL.md into your Claude Code skills directory:

~/.claude/skills/cloneify/SKILL.md

Then invoke it:

/cloneify

…or just describe wanting to clone something ("I want a local version of X", "let's clone Superhuman") and Claude will trigger it.

Run folder

All artifacts for a run live beside where you'll build, in your workspace — not inside the skill folder:

<run-root>/                 # e.g.  <cwd>/<slug>-local/  — the target repo / build dir
└── clone-run/              # process artifacts, beside the app code
    ├── STATE.md            # resume across sessions
    ├── scope.md            # Phase 0
    ├── research.md         # Phase 1
    ├── PLAN.md             # Phase 2
    ├── PLAN-REVIEW-LOG.md  # Phase 2 — the full Codex argument
    └── GOAL-PROMPT.md      # Phase 3 — the paste-ready build prompt

STATE.md makes runs resumable — re-invoking cloneify picks up from the last completed phase.

Example

You: /cloneify → "WisprFlow, but 100% local on Windows"

cloneify interviews you (platform, features, the ≤1.5s latency constraint, your GPU), deep-researches how WisprFlow works and the local-ASR landmines, grills you on the open decisions (which cleanup model, which ASR runtime, injection method), has Codex tear the plan apart until it's sound, and hands you one prompt that builds the whole thing — GPU-stack-first so the risky part fails fast, not last.

Credits & license

  • cloneify is released under the MIT License (see LICENSE).
  • Phase 2 builds on grill-me-codex, which in turn builds on Matt Pocock's grill-me (MIT). The grilling methodology lineage is his; the cross-model Codex review and the four-phase clone pipeline are additions.