Figma preflight
Read-only gate before an authorized canvas write. It returns clear preconditions or concrete gaps and never mutates the canvas. Read-only inspection does not require write approval. A gate pass or lock is not user consent; apply authorization already present in the task without asking again.
Preconditions for the affected scope
- Live connection and correct file: figma-console
figma_get_statuswith a probe and correct pairing, or the available official MCP identity + metadata path. Local process/port output alone cannot prove pairing. - Precise target id/selection and fresh structure. Resolve ambiguity only when it affects the target; valid metadata in context can be reused.
- Styled writes resolve the relevant variables/styles; new UI searches existing components/icons and app assets before inventing geometry. Report unrelated DS debt instead of turning a local edit into a file-wide cleanup.
- Read applicable canon/API details for this operation. Plan the product states and flow outcomes it actually affects; no universal manual bundle or required mental pass count.
- Claim the file lock using this skill’s
scripts/figma_lock.py, with explicit agent/task identity. Same-file single writer remains the invariant. A conflict, missing/broken lock script or uncertain owner keeps writes blocked. Never release another task’s lock, override autonomously on “disjoint pages”, or use an empty process snapshot as permission to bypass.
gate-checks.md explains tool alternatives, targets, variables, reuse and states when a check needs detail. file-lock.md documents the existing TTL lock CLI and release rules. Release only your exact agent/task lock in cleanup/finally. An explicit user exception must name what is waived; it does not waive every gate, fix missing access or authorize another task’s lock release by analogy.
Execution and proof
For partial errors/timeouts, read the canvas before retrying. Verify the property actually changed, not an unchanged proxy; preserve existing work and guard null nodes/live edits. write-verification.md contains the fragile Plugin API cases. Do not invent support for a tool/API.
New-screen or requested graph audit uses flow-audit.md: entry, destinations, orphan/dead/broken edges and semantic mismatch. An audit never creates a missing screen or rewires a valid flow on inference alone. Report the target, clear/gap status and evidence briefly; a tiny write does not require a ceremonial eight-row stamp or repeated user confirmation.
