Three well-defined tools with complete input/output schemas, clear descriptions, and proper parameter constraints. All tools use enums for constrained inputs (sources, categories) and include min/max bounds on numeric parameters. Tool annotations (readOnlyHint, destructiveHint, idempotentHint, openWorldHint) are present and correct. Descriptions are substantive (100-200 chars) and explain WHAT, WHEN, and WHAT-RETURNS. Parameter descriptions are present and actionable. Main gaps: no error handling guidance in descriptions, no recovery hints for edge cases (e.g., 'if no results found, try broader categories'), and output schemas lack per-field descriptions to guide LLM interpretation of importance scores and corroboration signals.

• Add field descriptions to digestOutputSchema output object. For each field (importanceScore, importanceReason, corroboration, latestPublishedAt), explain what it means and how LLMs should use it. Example: 'importanceScore: 0-100 heuristic ranking based on publisher prominence and recency; not editorial judgment.'
• Enhance tool descriptions with recovery hints. Example for get_news_summary: 'If results are sparse, try expanding lookbackHours or removing category filters. If a source fails, the warnings array will indicate which feeds were unavailable.'
• Document expected query formats in search_news. Clarify whether the query accepts phrases ('UK election'), single words ('election'), or boolean operators. Provide 1-2 examples.
• Add a 'warnings' field description to explain common warning types (e.g., 'partial-feed: RSS endpoint returned incomplete data', 'stale-cache: results may be older than requested lookbackHours').
• Consider adding a 'nextRefreshTime' field to output schema to guide LLMs on when to call refresh=true, improving cache-awareness.