A repository containing skills, agents, and validation scripts for an MCP-based system. This is not an MCP server itself, but rather a collection of skill and agent definitions with validation tooling.
Code Virtuoso defines 6 tools with partial schemas and descriptions. Most tools have descriptions (4-5 sentences), but parameter documentation is inconsistent. The annotate tool has 30+ parameters with minimal per-parameter descriptions, most lack type info beyond array/string/integer. Schemas are present but incomplete: many parameters lack descriptions, enums are missing for color/mode parameters, and output schemas are undocumented. Error handling and recovery guidance are absent. Tool names follow verb_noun convention (annotate, capture_*), which is good, but parameter naming is inconsistent (e.g., 'rect' vs 'circle' vs 'arrow' as arrays without clear coordinate semantics). No tool declares permissions, audit trails, or idempotency guarantees.
Annotate an image with shapes, arrows, text, numbered steps, blur, or redaction. Operations apply in order; later operations win for overlapping marks.
Capture a webpage screenshot using the system's native browser and screenshot tools. Cross-platform support for macOS, Linux, and Windows. Tier 3 fallback method.
Capture a webpage using Playwright in headed mode with stealth patches to bypass bot detection. Tier 1 anti-bot fallback method.
Capture a webpage by connecting to a user's already-running Chrome instance via Chrome DevTools Protocol. Tier 2 anti-bot fallback method using real browser session.
Copy a screenshot file from an MCP browser Docker container to the host filesystem. Resolves container-internal paths and handles auto-detection of the container.
Generic generator that fills data-explorer-template.html from a CSV file with domain-specific configuration for charts, stats, filters, and search.
annotate tool has 30+ parameters with minimal per-parameter descriptions. Most lack type constraints, enums, or format guidance. E.g., 'rect_color' accepts any string; should enumerate valid color names or hex patterns.
No output schemas documented for any tool. LLMs cannot infer what fields to expect or plan downstream calls. E.g., does annotate return {success: bool, output_path: string} or just a path string?
Parameter naming inconsistency: annotate uses 'rect', 'circle', 'arrow' as arrays without documenting coordinate order (x1,y1,x2,y2 vs center_x,center_y,radius). Descriptions say 'as x1,y1,x2,y2' but circle is 'center_x,center_y,radius', confusing for LLMs.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 46 | 2026-07-28+ | v2 |
No error handling or recovery guidance. Tools lack descriptions of failure modes, retryability, or next steps. E.g., what happens if the input image doesn't exist? Should the agent retry or ask the user?
No tool declares permissions, audit requirements, or idempotency. Destructive tools (copy_mcp_capture_to_host writes files; annotate modifies images) lack confirmation or dry-run support.