{"slug":"xuanranl-phase-publish","name":"phase-publish","description":"Run Phase Publish — image generation sub-pipeline (provider-pluggable: relay primary + official OpenAI fallback, realtime-forced) + WordPress publish with RankMath meta + IndexNow ping. Use when draft has passed all quality gates and is ready for live publication. Triggered by /seo-blog publish, /publish, \"post this to WordPress\", \"go live\".","long_description":"---\nname: phase-publish\ndescription: Run Phase Publish — image generation sub-pipeline (provider-pluggable: relay primary + official OpenAI fallback, realtime-forced) + WordPress publish with RankMath meta + IndexNow ping. Use when draft has passed all quality gates and is ready for live publication. Triggered by /seo-blog publish, /publish, \"post this to WordPress\", \"go live\".\nallowed-tools: [Read, Write, Bash, Task]\ndisable-model-invocation: true\n---\n\n# Phase Publish Orchestrator\n\nImage generation + WordPress upload + indexing notification + monitor registration.\n\n> **Critical**: `disable-model-invocation: true` — only triggered explicitly by user or upstream skill. Cannot be auto-invoked. Prevents accidental publishing.\n\n## Inputs\n\nRequired in `workspace/{task_id}/`:\n- `draft.md` with frontmatter `Stage: optimized`\n- `meta.json`\n- `quality.json` with `overall_passed: true`\n- `review.json` with `score >= target_score`\n\nRequired in `~/.xuanran-seo/credentials/wordpress/.json`:\n- `{ url, username, app_password }`\n\n## Stage 1: Image Sub-Pipeline (5 sub-stages, 1-34 min total)\n\n```\nStage 27a: image-slot-allocator (~2 sec, no LLM)\n Decide brief.image_count slots (default 6, max 8 — scripts/_core/image_policy.py): 1 cover (16:9) + the rest section images (16:9/4:3)\n Format-aware H2 selection (per BLOG-FORMATS-2026-CATALOG §2.2)\n Output: image_slots.json\n\nStage 27b: image-prompt-designer (~15 sec, LLM)\n Read brand-identity.json (colors), brand-voice.md (aesthetic)\n Generate shared Art Direction Prefix (Strategy A visual consistency)\n Design 9-field prompt per slot\n Output: image_prompts.json\n\nStage 27c: openai-image-generator (~1-3 min realtime, parallel max_workers=4)\n Routes to the provider chain (scripts/_core/image_provider.py):\n primary = openclawroot relay (realtime only — relays have NO Batch API)\n fallback = official OpenAI (auto-tried PER-SLOT if the relay fails)\n Calls images.generate per slot, saves PNGs, writes images.json.\n Mode is FORCED realtime (config.yaml :: image.default_mode). Do NOT use\n --mode auto/batch here: that bypasses the relay (batch is official-only) and\n reproduces the 2026-05-21 \"worst-of-both-worlds\" wait.\n\nStage 27d: batch-job-poller (SUPPLEMENTAL — only if an official-OpenAI\n batch was ever submitted; the canonical realtime flow does not need it)\n GET /v1/batches/{id} → status; download output_file when completed\n\nStage 27e: image-post-processor (~10 sec, no LLM)\n EXIF strip + WebP convert + srcset 3 sizes + compress <200KB + SEO filename\n Generate ImageObject schema fragment\n\nStage 27f: image-curator (REFERENCE-ONLY — no Stage() dispatches agents/image-curator.md; alt/caption polish happens in the scripted image pipeline + wp_publisher media fields)\n Polish alt text per image (≤125 chars, keyword-natural)\n Generate optional captions\n Replace [IMAGE-SLOT-N] placeholders in draft.md\n Inject ImageObject into schema @graph\n Update frontmatter Stage: images-injected\n```\n\n**Cost**: provider-dependent. At 4K high the ledger books ~$1.67/image → ~$10/article at the default 6 images (~$13.4 at the max 8). The relay (openclawroot) is token-billed and typically cheaper, but cost_ledger deliberately over-estimates it with the official per-image table (conservative cost guard, see scripts/_core/image_provider.py).\n\n## Stage 2: WordPress Publish (~30 sec)\n\n`scripts/wordpress/wp_publisher.py` (the canonical publisher — `scripts/publish/` holds\nonly change_log/gsc/indexnow helpers). Categories are **resolve-only** at publish: an\nunresolvable name aborts before any WP write (create deliberately at init via\n`scripts.wordpress.setup_categories`, or opt in with `--allow-create-categories`):\n\n```python\n1. Upload every generated image to Media Library (multipart)\n - cover → featured_media\n - sections → inline references in body\n - All srcset variants uploaded\n - REQUIRED 4 metadata fields per upload: title + alt_text + caption + description\n \n2. markdown_to_html(body) with anchor links resolved\n\n3. PROJECT CSS INJECTION (required — see HARD RULE below)\n IF projects/{slug}/brand/article-css.css exists:\n css = projects/{slug}/brand/article-css.min.css (prefer minified)\n wrapper_class = read leading selector from CSS file\n (internal name, typically \"{slug}-pillar\")\n # PUBLISHED name may differ: style_tokens rewrites it at the publish\n # boundary. Resolve with scripts._core.style_tokens, never hand-write.\n body_html = wrap_with_gutenberg_css_block(body_html, css, wrapper_class)\n # Append any trailing inline JSON-LD AFTER the closing wrapper\n \n4. POST /wp-json/wp/v2/posts (status: **draft** — DEFAULT, do NOT publish yet)\n - title, slug, content (now CSS-wrapped), excerpt, categories, tags\n - featured_media = cover media_id\n - `status: \"draft\"` unless user explicitly opted in to publish-on-create (see step 8)\n - Returns post_id + preview URL\n\n5. RankMath / Yoast meta via canonical bridge:\n - PREFERRED: POST /wp-json/wp/v2/posts/{id} with body {\"meta\": {rank_math_*: ...}}\n (works because the MU-plugin xuanran-rank-math-rest-bridge.php\n registers RankMath keys with REST schema)\n - DEPRECATED: POST /xuanran/v1/rank-math-bridge (route returns 404 — only GET exists now)\n - Note: rank_math_robots enum is [\"index\",\"noindex\",\"nofollow\",\"noarchive\",\"noimageindex\",\"nosnippet\"]\n — \"follow\" is NOT valid and will 400. Use [\"index\"] (follow is implicit default).\n\n6. Inline JSON-LD for FAQPage / ItemList / BreadcrumbList — these are NOT \n generated by RankMath's free tier; append as