Version 1.6.2
- New Generate with AI inside the editor — a "Generate with AI" button now appears next to the post excerpt, WooCommerce product short & long descriptions, and any custom meta fields you choose, right in the WordPress editor (Classic and block). Click it, review the AI draft, and insert it into the field. Nothing is saved until you save the post. Uses the same content model and API key as your other content features — no extra cost path.
- New Custom meta field allowlist — under MxChat → Content → Settings, list the meta keys of your own custom fields to give each one a Generate button in the editor. Off-by-key by default: only the fields you add get a button. The whole feature can be toggled off in one click.
- New Truncation guard on AI-generated posts — every REST `/generate-content` call now inspects the resulting body for length-floor, mid-word `<p>X</p>` endings, no-terminal-punctuation mid-sentence cuts, stranded opened `<h2>` / `<h3>` / `<li>` / `<p>` tags, and "N best ..." / "top N" promises that don't match the actual heading count. On a red-flag, the draft post is moved to the trash and the REST call returns HTTP 502 with a structured error (`ai_body_too_short`, `ai_body_truncated`, or `promised_count_mismatch`) so callers can decide whether to retry.
- New WP-CLI command `wp mxchat-adv scan-truncations` — backfill scanner that surfaces previously-generated posts (post_date >= 2026-04-01, body length 1.5K-9.5K chars) flagged by the same heuristics. Read-only diagnostic; no auto-repair. CSV / table / yaml output via `--format=`. AJAX endpoint at `wp-ajax/mxchat_adv_scan_truncations` exposes the same data to future admin UI.
- Fixed AI drafting features no longer return empty results on the newest reasoning models (the GPT-5 family). The "edit meta with AI" tools and FAQ answer drafting previously came back blank on those models because of an internal response-length cap they handle differently; the cap and request format now adapt to the model automatically.
- Fixed REST-generated comparison posts no longer ship with a stale "Updated 2024" / "Updated 2025" sidebar label. Generated post bodies are post-processed to rewrite the year string to the current year.
- Fixed Internal `MXCHAT_ADV_CONTENT_VERSION` constant could drift from the plugin header version (used a hardcoded literal). It now derives from the plugin header automatically, so future releases always emit the correct version string for cache-busting and update reporting (the `/wp-json/mxchat-adv/v1/health` endpoint's `plugin_version` field is now always in sync with the header).
- Improved Faster, cheaper AI drafting on GPT-5-family models — short drafting tasks (meta, FAQ answers, field generation) now request minimal reasoning effort, and all AI features share one provider routing layer so model fixes apply everywhere at once.
- Fixed AI content tools now work with Claude Opus 4.7, Opus 4.8, and Claude Fable 5. The content calendar generator and the in-editor tool actions previously errored — and the shared AI router returned blank output — when one of these newer Claude models was selected. They now return content correctly.
