Changelog

1.2.0 — 2026-10-07

• Adds one-shot conversation handoff for selected Keep it and repair-direction choices through the MCP Apps ui/message bridge, with the existing ChatGPT compatibility bridge as a fallback.
• Keeps unresolved findings unresolved and sends only explicit user choices back to the host model.
• Displays optional canonical Writing Assistant method IDs, registry version, repository revision, and editing mode when supplied by an integrating host.
• Keeps the standalone Diagnostic independently packageable; the shared widget does not become a second editorial reasoning system.
• Updates the local preview and regression suite for decision handoff.

Release notes

Repository integration

• Adds this product independently under products/writing-diagnostic/ in Writing_Assistant.
• Adds a reproducible standalone ZIP command and packaging boundary checks.
• Declares the product's CommonJS scope explicitly, alongside the existing app's ES modules.
• Adds repository commands and documentation for running and verifying the products separately.

1.1.0 — 2026-10-02

• Reject missing or malformed findings instead of displaying a false clean result.
• Enforce unique IDs, canonical primitives, quote occurrences, array limits, and non-overlapping spans.
• Align the skill's whole-passage instructions with the tool schema.
• Declare output schemas, explicit tool annotations, authentication behavior, and a narrow widget CSP.
• Support current MCP discovery/response fields and preserve legacy initialization.
• Correct Apps initialization and the direct window.openai.toolOutput compatibility path.
• Add individual primitive filters, accessible selection states, a working Keep it choice, and recovery messages.
• Add local regression checks, a sample preview, public review cases, and Vercel/container deployment files.

This source release is not a completed public submission. Its live endpoint, public policies/support, hosted recording, installed-host tests, and portal checks remain pending.