Goalify
Aboudjem/goalify
Escopeia, prepara e entrega um plano executável que sobrevive /clear
Ver no GitHubSkill para Claude Code que prepara tarefas complexas em um arquivo /goal autossuficiente, permitindo que uma sessão limpa execute e verifique toda a tarefa sem improviso ou desvio de contexto. Você escopeia o objetivo, e a skill pesquisa o repositório, tranca decisões críticas, então entrega um plano executável que sobrevive a /clear.
Instalação
npx convez-ai Aboudjem/goalify1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Refatorações grandes sem perda de contexto entre sessões
- Migrações de código com critérios de sucesso verificáveis
- Auditorias e features complexas com decisões pré-travadas
- Automação de tarefas técnicas que exigem múltiplas etapas verificadas
Veredito

Skill bem pensada que resolve um problema real em Claude Code: o desvio de contexto e a falta de garantia em tarefas longas. A abordagem de preparação + execução em sessão limpa é sólida e o suporte a verificação de critérios é útil. Recomendado para equipes que trabalham com refatorações e migrações.
README
A tradução para o português deste README ainda está na fila.
Set the goal. Trust the run.
You scope a big task in chat: a refactor, a migration, a feature, an audit. Then you /clear for a clean session and the plan is gone. The run improvises, drifts on decisions you never made, and you can't tell whether it actually finished.
goalify closes that gap. It does the prep while it still has your context: it reads and researches the repo, then locks the few real decisions. Next it writes a /goal file with the finish line defined in commands the run can check. You /clear and run that file; a fresh full-context session executes the job and tests as it goes, stopping only when every criterion passes. On success, it cleans up after itself.
> /goalify migrate our API from callbacks to async/await, keep tests green
goalify researches the repo, locks the few real decisions, and writes the run file:
~/acme/.goal/callbacks-to-async.md
then prints the two steps you run yourself, in a fresh session:
/clear
/goal ~/acme/.goal/callbacks-to-async.md
[!IMPORTANT] goalify prepares the run; it doesn't run your task here.
/goalifywrites the file; the/goal <path>it prints is what you run next, in a fresh session, and that file deletes itself once the run succeeds. Your plan survives/clear.
Quick Start — plugin install (recommended)
goalify ships in the 10x marketplace:
claude plugin marketplace add Aboudjem/10x
claude plugin install goalify@10x
Restart Claude Code if it's already open, then run:
/goalify migrate our API from callbacks to async/await, keep tests green
Or just say it:
goalify this: <your task>
goalify writes a /goal file and prints two steps to run in a fresh session:
/clear
/goal ~/your-repo/.goal/your-task.md
Install (manual / skill-only)
git clone https://github.com/Aboudjem/goalify
mkdir -p ~/.claude/skills
cp -r goalify/skills/goalify ~/.claude/skills/goalify
Restart Claude Code if it is already open so it loads the skill.
Two ways to start
Run the command:
/goalify <your task>
Or just say it, and the skill triggers on its own:
goalify this: <your task>
Either way, goalify researches, locks the few real decisions in one quick batch (skipped if there are none), and prints your two steps: /clear, then /goal <path>.
What you can trust
- Sourced, then re-checked. Every load-bearing fact in the file carries a source, and a separate agent re-derives the key claims.
- Knows when it's done. Success criteria are wired to real commands, so "finished" means the checks passed.
- Tested. Built test-first, with a recorded RED→GREEN baseline on Haiku, Sonnet, and Opus (baseline).
- Safe by design. Read-mostly prep, no remote fetch-and-execute, and a self-destruct that fires only on full success (security).
What goalify writes
A /goal file (see a real one) holds:
- a declarative spec: the end state and how it's verified, instead of a brittle step list;
- verified context with absolute paths, so the fresh session is never lost;
- machine-checkable success criteria tied to real commands;
- parallel fan-out for independent work, a progress checklist, and a gated self-destruct that deletes the file only on full success (and keeps it to resume otherwise).
FAQ
Does it run my task? No. It writes the /goal file; you run it after /clear, in a fresh session. That separation is the point.
Why a file instead of a prompt? A chat plan dies at /clear. A file persists, carries absolute paths and cited research, and the run re-reads it every loop.
What if the run can't finish? The self-destruct is gated. If any criterion is unmet, the file stays in place so you can resume from it.
Does it work outside Claude Code? It's a spec-correct Agent Skill, so it should be portable to agents that support the Agent Skills standard. The /clear and /goal hand-off is specific to Claude Code; adapt those two commands elsewhere.
When should I not use it? A one-line fix (just ask Claude), or open-ended exploration with no end state. goalify will decline rather than write a vague file.
Is there a plugin? Yes — goalify ships as a Claude Code plugin in the 10x marketplace (claude plugin install goalify@10x), and the skill still works standalone (copy it into ~/.claude/skills/). (Someone opened a Claude Code issue asking how to carry a plan across /clear; it was closed as not planned, so goalify is one answer.)
How it works
- Research & decide. goalify inspects the repo and locks the few real decisions.
- Write the
/goalfile. A self-contained, self-deleting run file at an absolute path. - Run in a fresh session. You
/clear, then/goal <path>; it executes, verifies, and cleans up.
The skill itself lives in skills/goalify/SKILL.md; the evals are in evals/; for a first run, see the quickstart.
Contributing & license
Issues and PRs are welcome, and goalify is built test-first (contributing · code of conduct). MIT.
Built by Adam Boudjemaa. Verified against Claude Code and the Agent Skills spec, 2026. Spot a gap?