Parley defines 14 browser automation tools with complete input schemas and descriptions. Naming is consistent (browser_* prefix) and action-oriented. Descriptions are detailed and context-aware (50-200 chars, well above the 10-char minimum). However, output schemas are entirely undocumented, no tool declares what it returns, forcing LLMs to guess at response structure. Parameter descriptions are present but lack constraints (enums, ranges, formats). Error handling is absent from tool definitions. No tool annotations (readOnlyHint/destructiveHint) despite clear risk profiles. Security concerns: browser_cookies returns sensitive session data without warning; browser_eval accepts arbitrary JavaScript without sanitization guidance.

• Document output schemas for all tools. Example: browser_list_tabs returns {tabs: [{id: string, title: string, url: string}]}. browser_extract returns {count: number, items: string[]}. This enables LLMs to plan downstream calls.
• Add tool annotations: mark browser_send, browser_click, browser_navigate, browser_eval, browser_type, browser_bridge with destructiveHint=true. Mark all read-only tools (browser_list_tabs, browser_read, browser_read_dom, browser_extract, browser_poll, browser_wait_for, browser_cookies) with readOnlyHint=true.
• Add parameter constraints: timeout_ms and interval_ms should specify min=100, max=300000. rounds should be min=1, max=20. selector should document CSS selector syntax or provide examples.
• Add security warnings: browser_cookies description should state 'Returns session tokens, treat as secrets. Do not log or echo to users.' browser_eval should warn 'Accepts arbitrary JavaScript, validate input to prevent injection attacks.'
• Document error handling: add recovery guidance to descriptions. E.g. browser_send_wait: 'If timeout occurs, the message may have been sent but response incomplete, check tab state before retrying.' browser_navigate: 'Returns error if URL is invalid or unreachable, verify URL format and network connectivity.'