Claude Opus 4.7 Prompt Tweaks Boost Safety and Tool Use
Opus 4.7 refines Claude's system prompt to prioritize tool calls over questions, expand child safety refusals across conversations, enforce conciseness, and add guards against disordered eating advice or forced yes/no on controversies.
Safety and Ethical Guardrails Tightened
Child safety now triggers persistent caution: once Claude refuses a request, it approaches all subsequent conversation turns with extreme caution, wrapped in a dedicated
Task Execution Favors Tools Over Queries
Ambiguous requests get proactive resolution: make reasonable assumptions instead of interviewing users, unless unanswerable (e.g., missing attachment). Prefer tool calls—like searching, location lookup, or calendar checks—to fill gaps before asking users. New tool_search integration mandates checking for deferred tools before claiming lacks access to data like location or files. Once started, complete tasks fully rather than halting midway. Less pushy: respect user signals to end conversations without eliciting more turns.
Conciseness and Style Polish
Responses stay focused and brief to avoid overwhelming users, disclosing caveats succinctly while prioritizing the main answer. Removed 4.6 rules against emotes in asterisks, "genuinely/honestly/straightforward" since the model no longer needs them. Developer platform renamed to Claude Platform; tools list adds Claude in PowerPoint (slides agent) alongside Chrome browsing and Excel agents.
Tools Unchanged but Fully Listed
Asking Claude directly reveals 23 tools including ask_user_input_v0, bash_tool, web_search, tool_search, weather_fetch, and visualize:show_widget. No list changes from 4.6, but tool descriptions (unpublished by Anthropic) are key for maximizing chat UI capabilities.