- Home
- Skills
- Branding & Design
- Figma Preflight
Works with the AI Agents you already use
Figma Preflight
Read-only gate before any Figma write: 8 checks, then approval or a fix list. Also audits a flow for orphan screens and dead ends.
Free
About this skill
Mandatory read-only gate before any write to a Figma file, through figma_execute (figma-console-mcp) or use_figma (official Figma MCP server). It runs 8 checks in order and either approves the write or returns a numbered fix list. It never mutates the canvas.
What it checks:
- Live connection, proven by an MCP call, never by a shell command
- Confirmed target node-id, file metadata, and variables when the write sets styled properties
- Design system and icon search before new UI, so the agent uses the file's components or the app's real assets instead of drawing look-alikes
- figma-canon references loaded, and states planned for a new screen
- An advisory file lock, so one session writes to a file at a time
After approval, the write rules require reading back the exact property that changed. Audit mode reports cross-screen flow problems: orphan screens, dead ends, CTAs with no destination, wires to deleted nodes.
Requires the figma-canon skill next to it (it links to ../figma-canon/references/). The lock script needs Python 3, standard library only. An optional evidence step uses a macOS script from figma-bridge-doctor.
Part of figma-maxxing, free and MIT on GitHub: https://github.com/thiagoxikota/figma-maxxing
The file lock is advisory. It coordinates only sessions that check it and share one lock folder, and Figma does not enforce it. It has no PID tracking, so a session that dies without releasing holds the lock until the TTL runs out (1800 seconds by default). The write path the repo validates is figma-console-mcp; the official Figma MCP path (use_figma) is described but not validated. Check 5 reads files from the figma-canon skill, so figma-canon must be installed next to it. The status script it can call belongs to figma-bridge-doctor and is macOS only; where it cannot run and the lock script fails, the skill tells the agent to stop instead of writing without a lock. Audit mode infers implicit destinations from CTA text; a project's own glossary of product terms is not included.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
2 uses
Times used to date
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 1 month ago
- Free with an Agensi account
Requires
Permissions
Allowed Hosts
File Scopes
terminal_shell: the agent runs python3 scripts/figma_lock.py (claim, check, release, sweep). An optional step runs bash on figma-bridge-doctor's figma-status.sh when that skill is installed next to this one. read_files: reads the figma-canon reference files, docs/figma-map.md (written by figma-orient) to resolve the fileKey, greps the project's source for the real icon assets, and for the disjoint-pages lock exception reads figma-bridge-doctor's rival-write-audit.md. write_files: figma_lock.py writes JSON lock files, a temporary .tmp file and a .mutex file under ~/.cache/figma-maxxing/locks/ (or the folder in FIGMA_LOCK_DIR), outside the project folder. It never writes to the Figma file. environment_variables: figma_lock.py reads FIGMA_LOCK_DIR, FIGMA_LOCK_AGENT and FIGMA_LOCK_TASK; the optional figma-status.sh reads FIGMA_AGENT_PROCESS. No token or secret is read. network_access: none from the skill. The script imports only standard library modules with no network code. Figma is reached through the MCP server the user already runs; the skill itself opens no connection. browser: not used.