Three tools with clear action verbs and reasonable descriptions (100-150 chars each). All parameters have type definitions and descriptions. However, output schemas are completely undocumented, LLMs cannot plan downstream calls or extract structured data. Error handling returns generic exception strings rather than actionable recovery guidance. No parameter validation constraints (enums, ranges) despite accepting free-form strings for 'focus' and 'recency'. Missing documentation of what fields are returned in the formatted response.

• Document output schema for all tools. Specify that response is a string containing formatted answer + citations section. Define citation format (e.g., '[1] URL' per line). Enable LLMs to extract and structure results.
• Convert 'focus' and 'recency' parameters to enums in schema. Define focus as enum: ["", "academic", "finance"] and recency as enum: ["", "day", "week", "month", "year"]. Remove free-form string acceptance.
• Replace generic error handling with categorized, actionable errors. E.g., 'API timeout (retryable): Perplexity API did not respond within 30s. Retry with a simpler query or try web_ask instead.' Include classification (retryable/user-fixable/fatal).
• Add confirmation step for deep_research. Return a structured preview (estimated duration, query summary) and require explicit confirmation before executing. Prevents accidental expensive calls.
• Document timeout behavior and retry strategy. Add to tool descriptions: 'If no response within Xs, the tool returns a timeout error. Retry with a shorter query or simpler focus.'