Critical gaps across all dimensions. 33 tools with minimal descriptions (most under 50 chars, many trivial). Input schemas visible for only ~8 tools (browser_click, browser_fill_form, browser_drag); remaining 25 tools have no visible schema definitions in source. Parameter descriptions are sparse or absent. No documented output schemas. Error handling and recovery guidance absent. Tool composition shows significant overlap (browser_* vs tabs_* duplicates same actions). No security annotations or permission gates visible. This server prioritizes feature breadth over definition quality, a pattern that causes LLM confusion and misuse.
Clicking page element
Waiting for download
Dragging page element
Filling form
Inspecting form semantics
Handling dialog
Hovering page element
Sending key input
25 of 33 tools have no visible input schema definitions in source code. Only browser_click, browser_fill_form, and browser_drag show explicit JSON Schema. Remaining tools (browser_hover, browser_type, browser_scroll, browser_key, all tabs_* tools, webmcp_*, domain_memory_*, skills_*, etc.) lack documented parameter types and constraints.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 31 | 2026-07-28+ | v2 |
Scrolling page
Capturing browser snapshot
Typing into page element
Preparing upload
Querying visual region
Waiting for page state
Deleting Domain Memory
Loading Domain Memory
Saving Domain Memory
Updating Domain Memory
Saving HTML App Artifact
Fetching URL
Loading skill details
Saving Domain Skill proposal
Capturing & querying VLM
Activating tab
Clicking page element
Creating tab
Filling form
Highlighting page element
Hovering page element
Typing into page element
Navigating tab
Discovering WebMCP tools
Invoking WebMCP tool
All 33 tool descriptions are trivial (10 - 50 chars). Examples: 'Clicking page element', 'Hovering page element', 'Typing into page element'. These lack context for LLM selection, no explanation of WHEN to use each tool, what it returns, or how it differs from similar tools. Baseline for A+ tools is 50 - 200 chars with clear intent guidance.
Severe tool composition overlap: browser_click, browser_hover, browser_type, browser_scroll, browser_key, browser_upload_file, browser_handle_dialog all have near-identical tabs_* counterparts (tabs_click, tabs_hover, tabs_type, etc.). LLMs cannot disambiguate when two tools do the same thing. Naming convention does not clarify the distinction (browser_ vs tabs_ is opaque).
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields (e.g., does browser_snapshot return element refs? What structure?). No guidance on pagination, result limits, or field availability.
No error handling or recovery guidance visible. No indication of which errors are retryable, which require user input, or which are fatal. domain_memory_delete is destructive but has no confirmation or dry-run pattern. Agents cannot self-correct on failure.