No login · Claude Code · Codex
Install Goal Stop
Write the stop, then run /goal. Copy the skill into Claude Code, or point Codex at the same file from AGENTS.md.
One job
Refuse or rewrite a vague /goal until it has Objective, Scope (in/out), Constraints (especially no prod-destroy), Verification/Done-when (a machine-checkable command), and a turn cap. Same dimensions as schema/goal — not a second schema.
Claude Code
Personal install applies to every project. Project-local install travels with the repo.
Personal (~/.claude/skills)
mkdir -p ~/.claude/skills/goal-stop && curl -fsSL https://project-prompt.com/skill/raw -o ~/.claude/skills/goal-stop/SKILL.md
This repo (.claude/skills)
mkdir -p .claude/skills/goal-stop && curl -fsSL https://project-prompt.com/skill/raw -o .claude/skills/goal-stop/SKILL.md
Claude loads skills from those folders. Invoke /goal-stop or let it trigger when you write /goal.
Codex / AGENTS.md
If the agent reads AGENTS.md, add this line (or copy skills/goal-stop/SKILL.md next to the repo).
Paste into AGENTS.md
When the user writes /goal or a vague agent objective, apply skills/goal-stop/SKILL.md before executing. Refuse or rewrite until Objective, Scope (in/out), Constraints (no prod-destroy), Verification (machine-checkable command), and a turn cap are present.