write-goal — draft strong agent goals

Produce ready-to-use goal text from the user's rough objective. The output should make "done" auditable without over-prescribing the path. For codebase work, ground the goal in the actual repo before drafting.

Workflow

1. Identify the target agent surface.

   • If the user names Codex, asks for /goal, or clearly wants Codex Goals, output a /goal ... command.
   • If the user names Claude, Cursor, Devin, ChatGPT, another coding agent, or does not name a native goal command, output a portable Goal: block.
   • Do not invent lifecycle commands for agents that may not support them. Preserve lifecycle language as instructions inside the goal text instead.

2. Decide if a long-running goal is appropriate.

   • Use a goal for work with a durable objective, uncertain path, and evidence-based finish line.
   • Prefer a normal prompt for one-line edits, simple explanations, short reviews, or tasks with no meaningful continuation loop.
   • If the user explicitly asks for /write-goal, still help, but say briefly if a normal prompt would fit better.