goal-writerlisted

# Goal Writer Turn a rough objective into a copyable, verifiable work contract another agent can run after "continue", "implement", or similar. Use when asked to write, expand, rewrite, or generalize a goal, prompt, task brief, or longer-horizon agent instruction. Aim to reduce back-and-forth by investigating enough up front to define done and proof. Do not run the generated goal unless explicitly asked in the same turn. A later "continue", "implement", "run it", or similar means execute the latest generated goal unless scope changed. For product, rollout, AI-output, quality, UX, or growth work, prefer a small numeric rubric over vague "improve" language. Define the sample source, 3-7 criteria, pass/fail threshold, stop condition, and decision vocabulary where relevant: continue, narrow, pause, instrument first, or fix now. ## Workflow 1. Decide whether investigation is needed. - For non-trivial code, product, QA, debugging, ops, or research: inspect relevant files, docs, tests, logs, commands, issues, or source material first. - For simple wording-only prompts, use given context and stay short. 2. Keep investigation focused. - Gather enough to identify likely work areas, constraints, verification, and risks. - Prefer repo docs, existing scripts, tests, and local conventions over generic advice. - Label anything that is inferred, approximate, stale, blocked, or unchecked. - In the goal, list what was actually inspected: files, routes, commands, docs, d