05-production

05 · Production — Draft Generation

This command orchestrates content-writer (vendored) plus Princeton GEO rewrite techniques and CITE/EEAT quality scoring via content-quality-auditor (vendored). It does NOT generate content from scratch — that work happened in 04-content-brief where the expert filled REQUIRED-FILL slots. Production assembles the filled brief into a polished draft and applies GEO-optimization rewrite rules.

---

Inputs

• clients/<slug>/brand_context.json (for voice + extended context)
• clients/<slug>/content_priorities.json (to pull the priority record)
• clients/<slug>/briefs/<id>.md (the expert-filled brief)
• clients/<slug>/briefs/<id>.meta.json (must have status: ready-for-production)

Output

• clients/<slug>/drafts/<id>.md — publishable Markdown draft.
• clients/<slug>/drafts/<id>.meta.json — machine-readable metadata (word count, citation count, statistics count, quote count, voice-match score, GEO-readiness checklist).
• clients/<slug>/drafts/<id>.html — interactive review version for the client (section-level comments + approve-as-is), rendered by the geo-review-html skill.

---

Procedure

Step 1 — Hard refusal gate

status=$(jq -r '.status' clients/<slug>/briefs/<id>.meta.json)
if [ "$status" != "ready-for-production" ]; then
  echo "REFUSE: brief <id> status=$status. Run 04-content-brief Step 9 first."
  exit 1
fi

If status is not ready-for-production, refuse to draft. Do not auto-fill slots. Send the user back to 04-content-brief.

Step 2 — Verify slot fills are substantive

Re-read the brief. For each former REQUIRED-FILL slot:

• Reject if content is <100 chars for original-data slots.
• Reject if expert-quote slot lacks attribution (name + role).
• Reject if customer-case slot lacks identifying detail (industry, approximate size, outcome metric).
• Flag (don't auto-reject) if content sounds AI-generated (hedging language, "in conclusion", "moreover", repeated bigrams).