sammyai

Project status

Last updated: see git history of this file.

Snapshot

• ✅ npm test passes – 218 unit specs exercise the background worker, analytics helpers, AI providers, and code generators.
• ✅ Integration coverage exists for the MCP JIRA→Figma bridge and AI provider glue.
• 🚧 Analytics backends (GA4, PostHog, custom webhook) still need real credentials before they will emit events beyond console logging.

Working features

• Story analysis (Figma → JIRA): background.js accepts screenshot + structured layer data, builds provider-specific prompts, and normalises Claude/OpenAI JSON responses into story objects before tracking success metrics. It also raises helpful configuration errors if the selected provider is missing credentials.
• JIRA ticket creation: the Atlassian REST client writes stories with Atlassian Document Format blocks, optional context, Figma deep-links, and platform labels, then returns the created issue key and browse URL.
• Figma layer capture: the worker rate-limits and caches calls to the Figma REST API, walks the node tree, and extracts dimensions, layout metadata, fills, strokes, effects, and text to enrich prompts when the plugin is unavailable.
• JIRA → Figma design generation: given a ticket payload, the worker downloads small attachments, generates Figma node JSON with GPT, and opens the configured Figma file with the plugin query string so the Figma plugin can build the frame.
• Code generation: the background worker validates rich design data, wraps the selected AI provider, and dispatches to React/SwiftUI/Kotlin generators in lib/codegen/. The generators compose prompts from the rich data and clean/validate the responses before surfacing them in the content script modal.
• Figma plugin bridge: figma-plugin/code.js can extract detailed design telemetry for story analysis and build frames from the bridge server’s plans. It communicates with the extension via postMessage and handles ready-state signalling.

Known gaps & risks

• Analytics backends are placeholders – without GA4/PostHog credentials or a custom webhook URL, events fall back to console logging only.
• Pricing tables and messaging inside the content script reference speculative GPT-5 costs; update them once real pricing is confirmed.
• Deployment automation docs were removed; recreate them once an actual CI pipeline exists.
• Code generation assumes rich design data – without plugin data it will error out. Consider a degraded mode that works from screenshot-only stories.

Recommended next steps

1. Configure real analytics credentials and confirm events land in the chosen backend.
2. Revisit the pricing/cost estimation logic in content.js once provider pricing is finalised.
3. Hook the deployment scripts into CI and document the release process in a dedicated ops guide.
4. Explore defensive fallbacks for code generation when only screenshot analysis is available.