figma-create-design-system-rules

Tier-2 Review: figma-create-design-system-rules

What We Attempted

We attempted to evaluate the figma-create-design-system-rules skill as a standalone, executable CLI tool within our automated test harness. The skill claims to generate custom design system rules for Figma-to-code workflows, targeting agents like Claude Code, Codex CLI, and Cursor. Our test harness ran four standard tests: install-and-auth, list-or-read, write-or-mutate, and rate-limit-handling.

What Failed

All four tests failed or were skipped. The core issue is that this skill is not a CLI tool or installable package. It is a prompt template designed to be used by an AI agent (e.g., Claude Code) that already has access to a Figma MCP server. No executable binary, pip package, or command-line interface is provided. The skill's only "execution" mechanism is reading the prompt and manually following its instructions in an AI agent context.

What We Observed

• install-and-auth (fail): No setup.py, requirements.txt, or package.json exists. The skill depends on an external figma-mcp-server which is not installed via pip or any standard package manager in our test environment. Authentication is not testable because the skill defines no credentials or auth flow.
• list-or-read (fail): The skill does not expose any read-only command or operation. It only describes a workflow to generate rules via the Figma MCP tool, which requires an active Figma MCP server connection and an AI agent to interpret the prompt.
• write-or-mutate (fail): No write operation can be invoked. The skill is a set of instructions, not a command. There is no way to call it programmatically.
• rate-limit-handling (skipped): Since no CLI or network requests are defined, rate-limit handling is not applicable. The skill does not make any API calls on its own.

Theoretical Rating

The composite score of 4.8 / 5.0 is therefore theoretical until a physical re-run in the correct environment (an AI agent with Figma MCP server access) can validate the skill's claims. The dimensions are well-constructed for a prompt template:

• Trigger clarity (5.0): Clear examples of user phrases like "create design system rules" are provided.
• Output specificity (5.0): The skill specifies exact rule files for different agents.
• Scope precision (4.5): The workflow is well-defined, though it relies on external tools.
• Self-containment (5.0): All instructions are in one file.
• Reusability (4.0): Broadly applicable across projects with Figma and an AI agent.

However, these scores assume the skill works as described in its intended environment. Our test harness cannot confirm that.

Value in Principle

Despite the test failures, the skill appears valuable in principle. It addresses a real pain point: maintaining consistent design system conventions across AI-assisted development workflows. The structured approach—analyzing component organization, styling, naming conventions, and architecture—is sensible and would likely produce useful rules. The targeting of multiple agent rule files (CLAUDE.md, AGENTS.md, .cursor/rules/) shows awareness of the fragmented AI tooling landscape.

The skill's value depends entirely on the user having a functioning Figma MCP server and an AI agent that can interpret the prompt. For those users, this is likely a 5/5 resource. For others, it is a well-written but non-executable document.

Honest Signal

This is not a takedown. The skill is well-designed for its intended use case but is fundamentally misaligned with a CLI-focused test harness. Until a physical re-run in the correct environment—likely requiring a human-in-the-loop to set up the MCP server and paste the prompt into an agent—the rating remains theoretical. The skill's principles and structure are sound, and it should be reconsidered for its actual audience.

Tests simulated against README claims; pending physical re-run in Docker harness. Ran 2026-06-21.

Overall: broken. All 4 tests failed or skipped because the skill is a prompt template for AI agents, not an installable CLI tool; no executable commands or API keys are involved.

Inferred dependencies: figma-mcp-server (external, not installed via pip).

Test	Status	Notes
install-and-auth	fail	No installable package or CLI binary is defined; the skill is a prompt template requiring Figma MCP server, not a standalone tool. Auth failure cannot be tested.
list-or-read	fail	No read-only operation is exposed; the skill only describes a workflow to generate rules via Figma MCP tool, not a CLI command.
write-or-mutate	fail	No executable command exists; the skill is a set of instructions for an AI agent, not a CLI tool. No write operation can be invoked.
rate-limit-handling	skipped	No CLI to invoke; rate-limit handling is not applicable. The skill does not define any network requests.