Scoring was not performed

• Document return schemas for all tools. For scrape_* tools, define: tweet object structure (text, author, metrics, urls), pagination fields (cursor, hasMore, total), and field types. Example: 'Returns { tweets: [{ id, text, author, likes, replies, urls }], cursor?, total? }'
• Add actionable error messages to tool implementations. Map API failures to guidance: 'Extension disconnected. Call get_extension_status() to verify, then retry.' Implement error classification (retryable, user-fixable, fatal) so agents can decide next step.
• Consolidate navigation + action patterns into single tools. Replace navigate_to_tweet → like_tweet with like_tweet(tweetUrl). This cuts call count, reduces context loss, and matches chat data model (users say 'like this tweet', not 'navigate then like').
• Add explicit constraints to media parameters. Define: accepted MIME types (image/png, image/jpeg, video/mp4), max file size (50MB for images, 500MB for videos), path normalization rules (expand ~/, resolve relative paths), and what happens if media is unreachable.
• Replace boolean draftOnly flags with explicit enum mode parameter. Example: mode: 'draft' | 'publish' | 'schedule'. Add field for schedule_time if schedule is an option. This removes ambiguity and lets agents choose intent explicitly.
• Add pagination support to scrape tools. Define limit (default 20, max 100), cursor/offset (for resuming), and return total count so agents can paginate through large result sets without blowing context.
• Implement dry-run or confirmation for destructive tools. post_tweet, post_thread, publish_article should support dryRun: true to show what would be posted without committing. This prevents accidental publishing to production accounts.