Three tools with complete input schemas and descriptions, but significant gaps in output documentation and error handling. Tool names lack action verbs (pplx_web, pplx_writing, pplx_custom are noun-heavy). Parameters are well-typed with enums for constrained fields (search_focus, source_focus, time_range, citation_mode), but descriptions are generic and lack actionable guidance. No documented output schemas, pagination support, or error recovery patterns. Parameter descriptions average ~40 chars, below the 72-char baseline for A+ tools. Missing guidance on when to use each tool variant and how outputs chain to downstream operations.

• Rename tools with action verbs: chat_with_perplexity_web, chat_with_perplexity_writing, chat_with_perplexity_custom. This signals intent and improves LLM tool selection.
• Document output schema for each tool. Include: response text, citations (format and structure), metadata (model used, tokens, search results count), and error fields. Example: {"response": string, "citations": [{"title": string, "url": string}], "model": string}.
• Add tool-level descriptions explaining when to use each variant. E.g., 'Use this tool for queries requiring current web information, news, or real-time data. Slower than writing mode but more accurate for time-sensitive questions.'
• Remove example values from parameter descriptions. Replace 'e.g. en-US, pt-BR' with a formal constraint: 'BCP-47 language tag (e.g., en, en-US, pt-BR). Defaults to en-US.' Let the schema enforce valid formats.
• Add error handling guidance to tool descriptions. Document: 'Returns error if model is unavailable, file attachment fails, or API rate limit is exceeded. Retry with exponential backoff for transient failures. Check model availability via Perplexity API before calling if allow_risky_model is false.'