0.0.7

[App 0.0.7] - 2026-04-30

Dual-mode desktop runtime. This release reshapes the app from a
Claude-dependent Bridge workflow into a product with two explicit
execution modes: MCP + Instructions and API + AI Chat.

Added - dual-mode workspace

• Home-screen execution mode selection. The user now confirms
  MCP + Instructions or API + AI Chat before choosing the analysis type;
  the selected mode is highlighted and persisted across restarts.
• Mode-specific setup flows. MCP + Instructions exposes a Claude
  Desktop guided setup and a Codex setup verification path, while
  API + AI Chat routes missing providers and models to Settings
  without forcing a home-screen connection check.
• Standalone native AI chat window. Native analysis opens the AI
  chat in a separate window instead of embedding it in the main
  workspace.
• Analysis status hero. Site analysis now keeps a visible mascot,
  animated progress stripe, and issue counters at the top of the main
  workspace during and after scans.
• Main analysis workspace. The main window keeps scan controls and
  structured analysis together, rendering confirmed facts, expert
  hypotheses, priority, expected impact, validation method, and scan
  totals.
• Second-screen details. Reports can be opened in a separate
  details window. Returning home leaves open details/chat windows in
  an inactive ended state.
• Report exports. The app can export generated audit reports as
  PDF, a standard Markdown document, or a lightweight HTML
  presentation through the same structured report contract.

Added - native runtime

• OpenRouter provider adapter. Replaced the Stage 1 stub with a
  real chat-completions request path, timeout handling, retry logic,
  and stable provider error mapping.
• Provider configuration. Added local provider setup with
  encrypted API-key storage through Electron safeStorage; raw keys
  do not round-trip back to the renderer.
• OpenRouter model profiles. One saved OpenRouter key can now
  power multiple saved model profiles, with a default model selected
  for analysis and follow-up chat.
• Structured report contract. Runtime responses must keep
  confirmed facts separate from expert hypotheses and include
  priority, expected impact, and validation guidance.
• Policy modes. strict_audit forbids hypotheses; audit_plus_ideas
  allows hypotheses only when explicitly labeled.

Changed

• Native mode no longer depends on Claude Desktop readiness. The
  legacy dependency gate remains relevant for Bridge Mode, but the
  native API path is intended to work with Claude Desktop fully closed.
• Bridge mode has no in-app AI chat. The integrated chat is scoped
  to API + AI Chat; MCP + Instructions keeps the live conversation in the
  external Claude Desktop workflow.
• Provider registry precedence. Environment overrides can win over
  stored provider config, which is useful for local testing and
  emergency recovery.
• Bridge scan state mapping. Bridge Mode state is mapped into the
  shared workspace so MCP facts can populate the analysis view without
  removing the Claude Desktop path.
• Standalone chat stability. The AI chat window no longer resets
  its own session in a render loop when the analysis report state is
  synchronized.
• Release workflow Node version. The app release workflow now uses
  Node.js 22, matching the root engines.node requirement.

Security and trust

• API keys are stored only in the Electron main process using OS-level
  encrypted storage when available.
• Provider errors are normalized before reaching the renderer.
• The UI and report contract keep factual tool output distinct from AI
  interpretation.
• Bridge prompt construction keeps the protocol token out of renderer
  prompt text; the token remains sourced from SKILL.md.

Verification required before tag

• Run the smoke tests in docs/SMOKE_TESTS.md.
• Confirm MCP + Instructions works end-to-end with Claude Desktop.
• Confirm API + AI Chat works end-to-end with a real OpenRouter key.
• Confirm separate chat/details windows, inactive states, and all
  export formats with realistic content.