One of my longer Claude 4.5 prompts:
create a spec at .claude/features/ for a statementStrategy model. (id, statement_id, written_by_sputnik_or_human, human_writer_user_id, approved_by_user_id, is_current_version (boolean), response_category (admit, deny, nonadmission), response_concise_title, response, response_reasoning). This feature lets the LLM create a strategy for each statement in a lawsuitEvent. We want individual creation (create for this statement), and then a Job to create all within this lawsuitEvent. Look in app/Ai and in Actions and in Livewire for patterns. For the UI, on the statement detail page, an empty strategy box is shown underneath the statement and its footnotes. 'STRATEGY' subtle title, on the right 3 dots with a flux dropdown: Generate, Write. This opens up a Strategy Flyout, that we will work out the details later, where a user can write or generate the strategy. (Also later, in /studio there will be a section where users can run tje Job to generate strategy for all statements in a lawsuitEvent). We keep multiple versions, and a field is_current_version (boolean) is used to know what version to display. In the flyout later users can go back to a previous version. Users can also write themselves. A user can also approve a specific strategy, then that is set as approved, until we un-approve it. So that a lawyer can review and approve the strategy. Go write a spec, and ask me questions. In the UI in the full card, we show 'strategy .. sputnik suggests/written by username; concise title (button to approve, response_category); response; reasoing with a dropdown to expand full reasoning (hidden by default). Show little layouts of the various card versions in the doc. Ask me clarifying questions.