content-flywheel

Content Flywheel

A repeatable, agent-driven content engine. One run produces a complete topic cluster: a primary (pillar) article, supporting cluster articles, platform social copy, and scheduled publishing — then analytics feed the next cycle.

Commands

Phases

Run phases in order. Each phase is best handled by the agent role noted — map these to whatever models/tools you have (a strong reasoning model for planning, a fast model for drafting, a research/grounding model for sources, a review model for the gate).

1. Research — trends, SERP intent, competitor gaps, audience needs. (research/grounding model)
2. Plan (2a) — write outline.md: H2/H3 framework, thesis, key points, internal-link plan, external-link candidates, CTA, frontmatter + structured-data plan. Inject [`frameworks/seo-writing-checklist.md`](frameworks/seo-writing-checklist.md) + [`frameworks/geo-techniques.md`](frameworks/geo-techniques.md) here. (strongest reasoning model — plans, does not write prose)
3. Draft (2b) — write the full article to the outline. (fast drafting model)
4. De-AI (2c) — strip filler, generic phrasing, and "AI tells"; tighten to the brand voice. (fast editing model)
5. Quality gate (2d) — score the draft against `frameworks/core-eeat-dimensions.md` + GEO rules; list gaps; fix before continuing. (review model)
6. Split — adapt the article into platform copy (see social limits below). (layout/copy model)
7. Carousel (optional) — HTML/CSS slides → image export.
8. Technical SEO — schema/JSON-LD, INP/LCP/CLS, alt text, slug. (technical model)
9. Publish — publish the article + schedule social. (automation)
10. Feedback — analytics → adjust content-pillar weights, cadence, and topic selection. Run this at the start of each weekly cycle so insights guide the new content. (reasoning model)

Trust score → autonomy level

A brand's trust_score.current (0–1) sets how much runs without a human:

Update trust after each human checkpoint: approved as-is +0.02; minor edits +0.01; major rewrite −0.05.

Cadence (optional cron)

• •Weekly — run Feedback (phase 10), then plan the next 7 days of topics into a weekly-plan.yaml (see schemas/weekly-plan.example.yaml).
• •Daily — read the plan, run phases 2–9 for that day's topic, catch up any missed day, then deploy.

Make runs idempotent: record a per-day ledger and exit cleanly if today is already done; resume partial runs from the last completed phase. After updating any rule, re-run the affected phases — never reuse stale prior output.

Social fan-out limits (validate before generating)

A social post without a backing article does not count as a flywheel deliverable.

Definition of done

1 run = 1 published topic cluster = pillar + cluster articles + platform copy + scheduled posts. Articles must pass the CORE-EEAT gate. Accumulate a batch of published items, then run Feedback to close the loop.

Files

• •schemas/brand.example.yaml — brand profile template (voice, audience, authors for E-E-A-T, URL rules, trust score).
• •schemas/content-matrix.example.yaml — content pillars + cluster topics.
• •schemas/weekly-plan.example.yaml — a 7-day plan.
• •frameworks/ — GEO, SEO, and CORE-EEAT references injected during planning and review.

Credits

CORE-EEAT / CITE frameworks adapted from aaron-he-zhu/seo-geo-claude-skills (Apache-2.0). Apache-2.0.