v0.2.10 - Windows PowerShell output visibility fix

Fixed

• Windows PowerShell output visibility: Child process output (Figma scan details, code generation progress) now displays correctly on Windows
  • Root cause: spawnSync with shell: true spawned cmd.exe as intermediary, which interfered with console output and ANSI color codes
  • Solution: Call Node.js scripts directly via process.execPath with argument arrays instead of shell command strings
  • Eliminates shell quoting/escaping issues and works identically across all platforms

v0.2.9 - Hybrid AST/regex validation

Changed

• Refactor validation layer to use hybrid AST/regex approach
  • Replace regex-based figma.*() extraction with ts-morph AST traversal for better edge case handling
  • Keep proven regex patterns for template/JSX validation
  • Improves correctness (+43%) and maintainability (+33%) with minimal performance impact

v0.2.8 - 4x faster Windows validation

Performance

• Replace npx with direct CLI invocation for Figma validation (4x faster on Windows)
  • Resolve @figma/code-connect binary path directly from node_modules
  • Use process.execPath (node) to invoke CLI instead of npx
  • Eliminates ~30s npx overhead per CLI call on Windows
  • Windows CI improved from 64s to 16s for unit tests
  • Reduce timeout from 120s to 30s (no package download overhead)

Changed

• Trigger CI unit tests on version tags in addition to branch pushes
• Improve release process documentation with pre-flight checks and error summaries
• Remove global Figma CLI pre-install from CI workflow (no longer needed)

v0.2.7 - Windows compatibility fix

Fixed

• CRITICAL: Windows compatibility for validation - add shell: true to all npx spawn calls
  • v0.2.6 used npx.cmd but still failed with EINVAL on Windows
  • Now uses shell: true which is the Node.js recommended approach for cross-platform compatibility
  • Removes platform-specific npx.cmd detection in favor of simpler, more robust solution

v0.2.6 - Windows npx compatibility fix

Fixed

• CRITICAL: Windows npx compatibility - use npx.cmd on Windows to prevent ENOENT errors
  • Previously caused 100% validation failure on Windows with "spawnSync npx ENOENT"
  • Affects all Code Connect file validation attempts
  • Now correctly detects Windows platform and uses npx.cmd

Changed

• Streamline agent documentation for clarity and reduced token usage

v0.2.5

🚨 CRITICAL BUG FIX

This release fixes a critical bug where @figma/code-connect was in devDependencies instead of dependencies, causing 100% validation failure for all global installs.

Fixed

• CRITICAL: Move @figma/code-connect from devDependencies to dependencies
  • Previously, global npm installs (npm install -g @bitovi/superconnect) did not include the Figma CLI
  • This caused 100% validation failure on all Code Connect files with "unknown error"
  • Now the CLI is always installed with superconnect
• Add upfront check for Figma CLI availability with clear troubleshooting steps
• Improve validation error messages to show exit code, stdout, and stderr for debugging
• Fix Anthropic SDK 0.71.2 compatibility:
  • Add explicit stream: false for non-streaming requests
  • Increase timeout to 20 minutes for long-running orientation tasks (SDK default was 10 minutes)

Changed

• Update dependencies to latest compatible versions:
  • @anthropic-ai/sdk: 0.18.0 → 0.71.2
  • commander: 12.1.0 → 14.0.2
  • openai: 4.104.0 → 6.14.0
  • undici: 6.18.1 → 7.16.0
  • @figma/code-connect: 1.3.12 (already current)
• Keep chalk at 4.1.2 and p-limit at 3.1.0 (newer versions are ESM-only, incompatible with CommonJS)

Full Changelog: v0.2.4...v0.2.5

v0.2.4

Fixed

• Add helpful error messages when API returns "Invalid model name" errors (400 status)
  • Shows current model being used
  • Suggests common alternatives (gpt-4o, claude-sonnet-4-5, etc.)
  • Explains how to set model via superconnect.toml or CLI flag
  • Links to model documentation
• Warn users when using custom base_url without explicitly setting a model
  • Prevents confusion when default model doesn't exist on custom endpoints
  • Applies to LiteLLM, Azure OpenAI, vLLM, and other OpenAI-compatible proxies

v0.2.3

Fixed

• Windows compatibility: Use Node's native fetch instead of undici to fix "fetch failed" errors on Windows PowerShell with corporate networks
• Include CHANGELOG.md in npm package files array so users can see release notes
• Change CI workflow to use npm ci instead of npm install for consistency with publish workflow
• Add .npmrc with engine-strict=true to enforce Node >=22 requirement (was only advisory before)
• Remove redundant .npmignore file (files[] array in package.json already controls what's published)

v0.2.2

Fixed

• Include .version-sha in npm package files array (was being generated but not included in published package, so --version only showed semver without git SHA)

v0.2.1

Added

• Include git SHA in npm package version output via prepublishOnly script
• E2E tests now run in parallel using GitHub Actions matrix strategy

Fixed

• Read api_key from superconnect.toml when validating agent token (interactive setup bug where custom API keys were written to TOML but not read during validation)