25 tools with Russian descriptions and schemas present, but significant quality gaps. All tools have descriptions (positive), but many are vague or lack actionable guidance. Parameter schemas are defined but descriptions are often minimal (1-2 sentences). No tool annotations (readOnlyHint/destructiveHint). Error handling returns JSON with 'error' and 'detail' fields but lacks recovery guidance. Output schemas are not documented, LLMs cannot predict response structure. Tools like 'agent', 'session', 'smart_scrape' combine multiple concerns. No pagination guidance for batch operations. Naming is generally clear (verb_noun pattern), but some tools are overloaded (e.g., 'session' with 6 action types).

• Add tool annotations: mark read-only tools with readOnlyHint=true, destructive tools (save_result, clear_cache, proxies) with destructiveHint=true, and idempotent tools with idempotentHint=true.
• Document output schemas for all tools. E.g., scrape returns {markdown?: string, html?: string, links?: string[], metadata?: object, ...}. Include this in tool docstrings so LLMs can plan downstream calls.
• Split 'session' tool into 5 separate tools: open_browser_session, click_element, fill_form_field, take_screenshot, close_browser_session. Each should have a single, clear responsibility.
• Enhance error messages with recovery guidance. E.g., instead of 'заблокировано антиботом', return 'Blocked by anti-bot. Try: (1) wait 10-30 min, (2) use smart_scrape with headless=false, (3) provide cdp_url for real Chrome session.'
• Expand parameter descriptions to 50-200 chars with context. E.g., 'formats: comma-separated output types. Use markdown for LLM-friendly text, html for structure, links to extract URLs, metadata for page title/description. Default: markdown.'.

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

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