Skip to content

Agent skills

cyber-figma ships workflow skills for Claude Code, Cursor, and other agents. Start here — skills encode when to reach for an MCP tool versus the CLI, how to keep a GET file call from burning a month’s Tier 1 quota, and the setup steps that turn a 403 into a fix.

The skills, the MCP server, and the CLI ship together as a plugin inside the cyber-figma npm package. Installing the plugin gets all three:

Terminal window
/plugin marketplace add cyberuni/cyber-figma
/plugin install cyber-figma@cyberuni

To take the skills on their own:

Terminal window
npx skills add cyberuni/cyber-figma

Or link individual skills into your agent’s skills directory — ~/.claude/skills/, ~/.cursor/skills/, and so on.

Set authentication before running any workflow.

Skill Use when
init-figma First-time setup — personal access token, team ID, connection verify, MCP wiring. Also the skill to reach for when commands fail with auth, permission, or “team id required” errors
inspect-figma-file Reading a file’s structure — pages, frames, components, styles, or one node — without wasting quota
export-figma-assets Getting images out — nodes rendered to PNG, SVG, JPG, or PDF, or the original image fills

The two file skills are deliberately split along the line the API draws: reading structure and rendering images are different endpoints on different rate-limit tiers, and conflating them is how an agent burns a Tier 1 budget.

Figma’s rate limits punish the obvious approach. GET file is Tier 1 — the most expensive tier — and returns the entire document tree with no pagination, while GET file meta is Tier 3 and answers most “what is this file” questions. On a View or Collab seat, Tier 1 is capped at roughly 6 calls per month on every plan, Enterprise included.

So the skills encode the cheap path first: metadata before documents, depth and ids before a full tree, and one batched GET images call listing many node IDs rather than one call per image — which is the mitigation Figma itself prescribes.