goal-to-planlisted

# Goal → Plan The bottleneck of an AI-native team is **not the model — it's the quality of the brief.** An agent earns autonomy (runs for hours, holds a real bar) once it has the four things a good new hire gets: a **clear goal**, the **right skills**, the **right tools**, and **rich, scoped context**. This skill turns a vague ask into exactly that — an artifact an agent (a lane, the runner, or a teammate's session) can pick up and run without re-briefing. It is a **skill chain**: it calls `premortem` (failure-modes → cheapest guard) and reuses the `sprint-scaffold` `spec_charter` template, then emits a dispatch block for whatever runs your agents (a lane, a runner, a teammate's session). Source idea: the Greg Isenberg / LCA "AI-native in under 60 minutes" planning pillar. ## When to use - A goal arrives fuzzy ("build the CRM", "fix onboarding", "do the affiliate program") and you're about to start work or dispatch a lane. - Before any multi-step / multi-session effort that should have one canonical, agent-readable home. - When triaging an inbound request (a call note, an issue, an email) into something actionable. ## Steps 1. **Compress the goal to ONE sentence + ONE acceptance criterion.** The acceptance criterion is the bar that makes "done" objective and testable ("a real call's card shows recent-calls exactly once", not "improve the card"). If the ask is genuinely ambiguous, ask **at most 2–3** targeted questions — otherwise pick sensible defaults, state them, and p