Vigil Risk exposes 6 financial data tools with clear action verbs (get_*, search_, read_) and domain-specific descriptions. However, critical gaps undermine production readiness: most tools lack documented output schemas, parameter descriptions are sparse or missing, and error handling is minimal. Tools are read-only (low risk) but lack the structured guidance LLMs need for reliable composition. The codebase shows defensive input sanitization (good), but tool definitions themselves are incomplete per the 54-pattern baseline.

• Document output schemas for all 6 tools. Example for get_stock_data: {type: 'object', properties: {ticker: string, current_price: number, pct_7d: number, pe_ratio: number, ...}, required: [ticker, current_price]}. This enables LLMs to plan downstream calls.
• Add WHEN-to-use guidance to tool descriptions. E.g., get_stock_data: 'Use this to fetch fundamentals for a single ticker before analyzing risk. Use get_market_pulse for macro context instead.' This prevents redundant calls.
• Implement structured error responses with recovery hints. E.g., if ticker is invalid, return {valid: false, error: 'Ticker INVALID not found. Try search_company_docs() to find the correct symbol.'} instead of plain text.
• Add limit and offset parameters to search_company_docs and read_company_doc. Cap results at 20 snippets for search, 12,000 chars for read (already in code as _DOC_READ_CAP). Document these limits in tool descriptions.
• Add parameter descriptions to get_market_pulse and get_sector_performance. Even though they take no inputs, document what they return: 'Returns VIX, S&P 500 trend, 10Y yield, gold, USD index, market regime, and fear/greed index. Call this first to assess macro backdrop before analyzing individual stocks.'

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry