Session Handoff — GEO/SEO/AEO stack audit + Zubair's SEO toolkit installed; next = lead-magnet website on her own domain

WHERE WE JUST WERE — verbatim (MANDATORY, always first)

Most recent assistant message (answering how to fill the DataForSEO signup wizard):

> Easy — these just tailor their onboarding tips; they don't change pricing or what your toolkit can pull. Answer them to match what /seo actually does: > > How will you use DataForSEO? (multi-select — pick the ones the toolkit uses): > - ✅ Keyword Research > - ✅ SERP Tracking > - ✅ Rank Tracking > - ✅ Backlink Analysis > - ✅ Competitor Analysis > - ✅ Content Optimization > - ✅ Local SEO (your /seo prospect finds local businesses) > - Optional: AI / LLM Mentions fits your AI-visibility angle — add it or skip, the toolkit doesn't need it. > - Skip: Merchant/Shopping, News/Media, Data Reselling, Building SaaS Tool. > > Primary integration method: REST API ← pick this one. Your toolkit talks straight to their REST API (not the SDK, not No-code). > > Describe your project: Optional — skip it, or paste this: > > Claude Code skill calling the DataForSEO REST API for client SEO audits — keyword research, rank tracking, backlinks, competitor and technical analysis. > > Then hit Continue and finish signup. > > The one part that actually matters comes after: go to app.dataforseo.com/api-access and copy your API password — it's a long random string, NOT your account login password (that mix-up is the #1 cause of the toolkit throwing a 401). Send me your login email + that API password and I'll wire them into the skill and run a live /seo audit to prove the whole thing end-to-end. > > ⏱️ Heads-up: this session is at ~285K tokens, past your 250K freshness line. Once we've got DataForSEO connected, it's a good moment to run /session-handoff.

Prior substantive message (install-complete report):

> ✅ Done — Zubair's SEO toolkit is installed, Windows-fixed, and verified (the safe way: cloned + ran locally, never curl | bash). > - 15 /seo sub-skills + 5 subagents registered · dataforseo_client bundled · reportlab in the venv (PDFs work) · shebangs re-pinned to the Windows venv Python so /seo runs · a real script executed and made a live DataForSEO call — it only stopped on the placeholder credentials. > - The one thing I couldn't test: a real audit with real data — that needs her free DataForSEO signup. > - Three pillars now: GEO ✅ strong (geo skill) · SEO ✅✅ upgraded, real Google data (dataforseo-claude) · AEO ✅ (geo-llmstxt/schema + AEO File Generator). > - Two open loops: (1) her free DataForSEO signup; (2) rotate the GitHub PAT flagged earlier. > - Recorded in ACCESS-MAP + memory + HISTORY (not committed yet). > - Next: the public lead-magnet website on her own domain that wraps this engine. Offered to write the PRD + Fable build prompt.

ACTIVE MODES — RE-INVOKE THESE FIRST (MANDATORY)

• fable-mode — five-gate discipline was governing this whole session (she invoked "/fablemode" explicitly to run the audit). Re-invoke it FIRST before touching the work. Does NOT survive /clear.
• ponytail — full level, active via session hook (laziness/YAGNI discipline). Auto-loads via hook, but confirm it's on.
• Standing writing rule if any copy gets written: her voice (context/voice/sondra-real-skool-posts.md), navy/gold/cream default theme.

Where it started

Sondra wants to turn her GEO tooling into an actual working website on her own domain — a free lead magnet (paste URL → score → capture email → funnel to her) modeled on Zubair's citevue.com and Sandy Lee's sleegeo.com. She was frustrated when I over-thought "forking" — she already downloaded + implemented Zubair's GEO engine. Real ask: "Do I have good GEO, SEO, and AEO — and what's missing?"

Decisions locked + what shipped

• Coverage audit: all 3 pillars owned. GEO = /geo (Zubair's 15-sub-skill suite, the engine CiteVue runs on); SEO = /seo (dataforseo-claude, installed this session, real Google data); AEO = geo-llmstxt/geo-schema + AEO File Generator + SEO-AEO Readiness Lab (both Base44 apps). Confirmed via disk sweep that NO standalone CiteVue/geo-seo web app exists — what she downloaded from his Git IS the skills.
• Installed Zubair's dataforseo-claude SEO toolkit — vetted 🟢 (MIT, 126★, clean installer), the SAFE way: git clone to scratchpad + local bash install.sh, never curl|bash. Landed at ~/.claude/skills/seo/.
• Fixed the Windows install break — installer assumes Unix venv (.venv/bin/python3); on Windows it's .venv/Scripts/python.exe. Installed reportlab into the venv + re-pinned all scripts/*.py shebangs to the venv python. Verified a script runs + makes a real DataForSEO call.
• Not forking/rebuilding — the build is a website wrapper around the engine she owns, on her own domain, NOT Base44.
• Recorded in context/ACCESS-MAP.md, HISTORY.md, memory project_geo_seo_aeo_stack.md.

Key files for next session

• C:\Users\srver\AIOS-Workspace\aios-starter-kit\PROJECTS\GEO-Agency\self-serve-app-spec.md — the May blueprint for the self-serve web app (read first; update to her current stack).
• C:\Users\srver\AIOS-Workspace\aios-starter-kit\PROJECTS\GEO-Agency\README.md — the agency playbook + competitive landscape (CiteVue/SleeGEO).
• C:\Users\srver\.claude\skills\seo\SKILL.md — the installed SEO toolkit (line 146-147 = the "call scripts directly via pinned venv shebang" note).
• C:\Users\srver\.claude\skills\geo\SKILL.md — the GEO engine (6-category scoring, 5 subagents, PDF gen).
• C:\Users\srver\.claude\skills\seo\.env — placeholder DataForSEO creds; drop her real login + API password here.
• Memory: C:\Users\srver\.claude\projects\c--Users-srver-AIOS-Workspace-aios-starter-kit\memory\project_geo_seo_aeo_stack.md
• Session log: C:\Users\srver\AIOS-Workspace\aios-starter-kit\outputs\session-logs\2026-07-11-2157.md

Running state

• Background processes: none (the repo-hunt shell b3wq8r1b8 already completed — empty result, nothing to kill).
• Dev servers / ports: none.
• Open worktrees / branches: none. (Cloned repo sits in the session scratchpad: ...\scratchpad\dataforseo-claude — disposable.)

Verification — how to confirm things still work

• ls ~/.claude/skills/seo/scripts/*.py — the 8 scripts are present (backlinks, domain_overview, keyword_research, on_page_audit, serp_check, prospect_finder, generate_pdf_report, dataforseo_client).
• head -1 ~/.claude/skills/seo/scripts/domain_overview.py — shebang = #!/c/Users/srver/.claude/skills/seo/.venv/Scripts/python.exe (the Windows fix).
• ~/.claude/skills/seo/.venv/Scripts/python.exe -c "import reportlab; print(reportlab.Version)" — prints 5.0.0.
• Once creds are in .env: /seo quick stripe.com — should return real Google data (proves the whole chain).

Deferred + open questions

• Open — DataForSEO signup (her step): she was mid-wizard; needs login email + API password (app.dataforseo.com/api-access, NOT the account password) → then run a live /seo audit.
• Open — The website is not built: wrap GEO+SEO+AEO into a public site on her stack (React+Vite + Cloudflare Pages + Supabase + Stripe + Resend), NOT Base44; needs a server-side audit runner (Worker + Claude API) so visitors get a ~30s result without her at Claude Code. Options A/B/C/D discussed; recommendation = Option B. Next = PRD + Fable build prompt (offered, not written).
• Deferred — HTML one-pager of the coverage audit for the CEO Dashboard (offered, not built).
• Open — rotate the GitHub PAT (plaintext in .git/config, printed to transcript).
• Open — quarterly-refresh checklist baked into the tool (her explicit requirement).

Pick up here

Re-invoke fable-mode first. Then either (a) if she has DataForSEO creds, drop them into ~/.claude/skills/seo/.env and run a live /seo audit to prove the toolkit end-to-end, or (b) write the PRD + Fable build prompt for the lead-magnet website (her stack + own domain, reusing GEO+SEO+AEO, with the quarterly-refresh built in).