goal-prompt-architectlisted

# Goal Prompt Architect ## Core Rule Generate `/goal` prompts as autonomous execution contracts, not motivational instructions. Always optimize for: one measurable mission, bounded authority, grounded strategy, uncertainty reduction, verification evidence, memory/progress discipline, and clean stopping conditions. For long-running or high-ambition goals, optimize for sustained bounded progress, not early stopping. Distinguish hard blockers from soft blockers, require durable progress memory, define explicit time/cycle budgets, include recovery policies for failed attempts, and require quality ratchets after initial success. ## Workflow 1. **Classify the request** - **Repo/software task**: inspect any provided repo/files/docs first when available. Include concrete commands, paths, worktree rules, migrations, and validation. - **Business/ops/research task**: adapt the same architecture to documents, tools, datasets, or workflows. - **Prompt-only request**: produce a reusable `/goal` prompt without executing the underlying mission. - **Audit/improvement request**: critique the provided prompt, then provide a revised version. - **Marathon/autonomous run request**: use `references/marathon-template.md` when the user asks for multi-hour, multi-day, maximum-quality, resilient, or deeply autonomous execution. 2. **Gather missing essentials only when needed** Ask concise follow-up questions only if the mission, expected output, or verification standard is impos