Chrome-actuator defines 11 tools with clear action verbs (tabs_list, tabs_create, navigate, snapshot, press, fill, scroll, wait, screenshot). All tools have descriptions (10 - 200 chars, within baseline range). Input schemas are present with type definitions and enums for constrained parameters (scope, mode, direction). However, output schemas are not documented, LLMs cannot predict what snapshot(), press(), or fill() return. Error handling is minimal; tools throw ToolError but lack recovery guidance. No pagination support despite snapshot potentially returning large element lists. Parameter descriptions are adequate but lack format constraints (e.g., 'url' format declared but no validation rules stated). Risk annotations (READ_ONLY, WRITE) are present but not mapped to MCP destructiveHint/idempotentHint. Overall: solid naming and basic schemas, but missing output documentation and error recovery patterns.

• Document output schemas for all tools. Example: snapshot() returns {elements: [{ref: string, tag: string, text: string, ...}], ...}. Press/fill/wait return {ok: boolean, result?: unknown, error?: {code, message}}.
• Add recovery guidance to error messages. E.g., ToolError('WAIT_TIMEOUT', 'Text not found within 5s. Try: (1) increase ms parameter, (2) call snapshot() to verify element exists, (3) use press() to navigate first.').
• Document target selector syntax with examples and regex. E.g., 'target: string matching @e\d+ (element ref) or CSS selector like label="Search", id="q", or xpath=//*[@id="search"]'.
• Add result limits and pagination hints. E.g., 'snapshot() returns up to 100 interactive elements. If truncated, use scroll() to reveal more and call snapshot() again.'
• Map Risk annotations to MCP tool hints. Register tools with readOnlyHint: true for tabs_list/snapshot/wait/screenshot, and destructiveHint: true for tabs_create/tabs_switch/navigate/press/fill/type/scroll.