fieldnote-ops/frameevidence

README excerpt

FrameEvidence Bounded, read-only design evidence for agent harnesses. FrameEvidence lets an agent inspect layout, typography, paints, component references, and variable bindings through the Figma REST API before it writes code. Version 0.1 is packaged as a plugin for DeepSeek Harness. FrameEvidence is deliberately read-only. Version 0.1 provides two tools: - figma inspect returns a bounded, implementation-focused node tree instead of dumping Figma's full JSON schema into the model context. - figma render returns a temporary PNG, JPG, SVG, or PDF render URL for one node. Evidence at a glance Surface Current boundary Figma access Read-only REST requests to https://api.figma.com/v1 ; redirects are rejected and no tool writes to Figma. Credential handling The PAT is read from a named host environment variable, never accepted as a model tool argument, and never returned in tool or probe output. Context control Raw responses are byte-capped; depth and returned node count are bounded before design evidence reaches the model. Maintainer infrastructure No maintainer server, analytics, telemetry, OAuth broker, or credential store is involved. Verified today 15 credential-free tests and clean…

Category