SKILL.md
FLOW: Find, Leverage, Optimize, Win
FLOW is an operating loop, not a checklist. Diagnose which stage the site is actually in, work that stage with evidence, and only advance when the stage's exit condition is met. Framework concept credit: FLOW by Daniel Agrici (CC BY 4.0). State this attribution once when the framework drives an engagement.
Stage routing
| Stage | Question it answers | Core work | Hand off to |
|---|---|---|---|
| Find | Where is the demand we are not capturing? | Keyword research, gap analysis, SERP intent mapping | claude-seo-cluster for clustering, claude-seo-content-brief for briefs |
| Leverage | What authority do we have and need? | Backlink and mention strategy, off-site authority | claude-seo-backlinks for link data |
| Optimize | Why do existing pages underperform? | On-page, technical, content, and AI-surface fixes | claude-seo-page, claude-seo-technical, claude-seo-geo |
| Win | Do rankings turn into revenue? | BOFU pages, conversion paths, dual-surface (classic + AI) scoring | claude-seo-sxo for persona scoring |
| Local (parallel track) | Do we win the map and the local SERP? | GBP, citations, local pages | claude-seo-local, claude-seo-maps |
Workflow
- Diagnose the stage. Ask what evidence exists: keyword coverage data (Find), link and mention profile (Leverage), pages ranking 5-20 (Optimize), rankings without conversions (Win). No evidence at all means start at Find.
- State the stage and why. One sentence, name the evidence.
- Work the stage. Pick 2-3 concrete analyses, not everything. For Optimize, select by context: SaaS leans on-page plus technical, local businesses lean citations plus GBP, publishers lean E-E-A-T plus freshness. Product pages point to conversion work, blogs to freshness and authority. Say which analyses you chose and why.
- Set the exit condition. Example: Find exits when a prioritized keyword-to-page map exists. Optimize exits when the flagged pages have shipped fixes and a re-crawl confirms them.
- Loop. Re-diagnose after each stage. Sites regress stages after redesigns and migrations.
Evidence rule
Every recommendation names its evidence: a SERP observation, a crawl finding, a data pull, or a documented platform behavior. No evidence, no recommendation. If live data would settle a question, use claude-seo-dataforseo or claude-seo-google before opining.
Verification
At the end of a FLOW session, list each recommendation with its evidence source in two columns. Expect zero rows with an empty evidence cell. Any empty cell: delete the recommendation or go collect the evidence.
Good vs bad
Good: "You rank 6-12 for 40 commercial keywords (ranked-keywords pull attached). That is an Optimize-stage profile. Running on-page analysis on the top 10 URLs by impressions." Bad: "Let's do keyword research, link building, on-page fixes, and CRO all at once." No stage diagnosis, effort spreads thin, nothing exits.
Footguns
- Starting every engagement at Find. Established sites with ranking pages usually need Optimize or Win first. Diagnose from data, not habit.
- Dumping every possible analysis at once. The Optimize stage alone could spawn a dozen audits. Surface 2-3, chosen by vertical and prior findings, and name the rest as available.
- Skipping the exit condition. Stages without exit criteria never end and never compound. Write the exit condition before starting the work.
Completion checklist
- [ ] Stage diagnosed and justified with named evidence
- [ ] 2-3 analyses chosen for the stage, choice explained
- [ ] Every recommendation has an evidence source
- [ ] Exit condition written for the current stage
- [ ] Correct sibling skills recommended for deep dives
- [ ] Attribution stated once for the engagement
Any box unchecked: not done. Fix or say so.