adapto-content-create

adapto:content-create

The third pipeline step (content-pipeline.md). It executes the briefs from the cycle plan into complete Markdown drafts — on-brand, with SEO/AEO/GEO metadata, JSON-LD, and internal links baked into the frontmatter — for you to review as files before anything reaches Adapto. It uses adapto-writer to draft and adapto-editor to critique, and writes nothing to the CMS.

Autonomous-safe (studio.md §4): local, no CMS writes — an autonomous cycle can run this unattended; the cycle parks at the human gate before adapto:content-upload.

When to use

• "Write the planned content", "draft these pieces", "create the articles from the plan".
• After adapto:content-plan has produced briefs (status briefed in the ledger).

When not to use

• Choosing what to write → adapto:content-plan.
• Pushing approved drafts to Adapto → adapto:content-upload.
• Translating existing content → adapto:translate.

Inputs

• The approved briefs in the latest .adapto/plans/<date>-cycle.md (and their ledger rows).
• The brain — voice.md, glossary.md, identity.md, audience.md, inventory.md (for internal links).
• The standards — seo-standards.md, prose-standards.md (the anti-slop rules the editor hard-gates on), and the frontmatter contract (content-pipeline.md §2).

Outputs

• A dated Markdown draft per piece at .adapto/drafts/<YYYY-MM-DD>-<slug>.md with full frontmatter (the adapto: + seo: blocks) and an on-brand body — the artifact you review.
• Ledger rows advanced to status: drafted (with draft_path).
• Next step: review the drafts, then adapto:content-upload to push the ones you approve to Adapto (as drafts).

Creation (write → critique → revise)

1. Select the briefs to write this run (default: all briefed in the latest cycle; or a subset the user names).
2. Write each brief — dispatch adapto-writer (one per brief). For briefs marked cornerstone: true, dispatch the writer at the top model tier (Opus-class); otherwise Sonnet-class (sub-agents.md §7). The writer reads the brain + standards and produces the md draft with full frontmatter.
3. Slop gate (deterministic) — before critiquing, grep each draft body for em dashes: awk '/^---$/{n++;next} n>=2' <draft> | grep -c '—' || true must return 0 (prose-standards.md — en dashes in numeric ranges are fine). If it's non-zero, hand the draft straight back to the writer to strip the em dashes (and the other AI tells) before spending an editor pass. No draft advances to drafted with em dashes in the body.
4. Critique — dispatch adapto-editor on each draft (voice fit, SEO/AEO/GEO completeness, internal-link coverage, structure, and the slop check — any prose-standards.md ban-list hit forces revise). If the verdict is revise, hand the gaps back to the writer and revise until solid.
5. Write the drafts to .adapto/drafts/, advance the ledger to drafted, and tell the user which files to read (perfectly named + dated). Point to adapto:content-upload for the ones they approve. A first draft is a new file — but when revising a draft that already exists, read it first (conventions.md §15): the harness blocks a write to an unread file, and the user may have edited it since.