Session Handoff — Migrate Claude Desktop (Claude.ai) projects + conversations into Claude Code

_Saved: 2026-06-29_

Where it started

Sondra wanted everything from her Claude.ai (Claude Desktop) account — all projects + conversations — pulled into Claude Code, organized, and permanently findable ("find it without a shadow of a doubt"), then surfaced on her CEO Dashboard. Constraints that emerged: she's voice-first/non-technical, reads HTML not markdown, and — critically — when she shows a reference page she likes, replicate it EXACTLY; do not redesign (she got furious when the dashboard page kept getting re-fonted instead of matching the MASTER-INDEX she pointed to).

Decisions locked + what shipped

• Approach: official Claude.ai export for content; Claude's own data API for the project labels the export strips. Playwright scraping rejected as the primary method.
• Export parsed: data-...batch-0000.zip -> 577 conversations, 28 projects (18 with content). She confirmed only one export email (no other batches).
• Key discovery: the export does NOT tag conversations with their project. Recovered the mapping via GET /api/organizations/{org}/chat_conversations (returns project_uuid per conversation) — driven through a logged-in Playwright browser.
• Rebuilt everything: 238 conversations filed under their 18 real projects + 339 loose filed + 49 recovered-from-deleted = 577 total, full content, zero missing. Lives in C:\Users\srver\AIOS-Workspace\aios-starter-kit\PROJECTS\Brain Migration\_projects-rebuilt\.
• Findability layer: grep-able INDEX.md, searchable conversation-catalog.html, and a two-pane _library/ app (sidebar A-Z/Recent/drag-reorder, resizable divider, persisted search/category, scoped search, ?cat= deep-link, 577 readable per-conversation pages).
• CEO Dashboard: claude-desktop-projects.html built to EXACT MASTER-INDEX styling (navy #0d1b2a header, Georgia gold #c9a227 h1, bordered alternating-row table, #8a6d00 links — do NOT redesign). Library copied to site/claude-desktop/library/. Home button "Claude Desktop Projects" added to both build/home-base.template.html and site/index.html.
• SOPs library created at SOPs/ with the first SOP + an HTML generator.
• NOT deployed yet — dashboard changes are local only.

Key files for next session

• PROJECTS\Brain Migration\ — all migration scripts: parse_export.py, file_all_projects.py, build_master_index.py, build_library.py, match_project_conversations.py, file_project.py
• PROJECTS\Brain Migration\_matches\all_conversations_mapping.json — the project->conversation map from Claude's API (THE key artifact; re-pull this to do a future re-sweep)
• PROJECTS\Brain Migration\INDEX.md — agent-facing master index (grep it to find any past conversation)
• PROJECTS\Brain Migration\_projects-rebuilt\MASTER-INDEX.html — the exact style Sondra approved; reference for any related page
• ceo-dashboard\build\build_claude_projects.py — builds the dashboard page (matches MASTER-INDEX; do not restyle). NOTE: NOT wired into deploy.bat — must run it manually before deploying.
• SOPs\sop-migrate-claude-desktop-to-claude-code.md + SOPs\build_sops.py
• Memory files touched: feedback_readable_html_large_high_contrast.md, feedback_offer_sop_after_notable_work.md, and MEMORY.md
• No plan file drove this session.

Running state

• Background processes: none
• A Playwright MCP browser is open and logged into claude.ai (used for the API pull). Sondra was told she can close it. No shell ID; close via browser or it idles out.
• Dev servers / ports: none
• Open worktrees / branches: none (on main; nothing committed this session)

Verification — how to confirm things still work

• cd "PROJECTS\Brain Migration" && python -X utf8 build_master_index.py — prints indexed 577 conversations / loose filed: 339 / archived: 49
• grep -i "veronica" "PROJECTS\Brain Migration\INDEX.md" — returns the Veronica conversations (retrieval works)
• Open ceo-dashboard\site\claude-desktop-projects.html in Edge — must look identical to MASTER-INDEX (clean table)
• Counts reconcile: 238 (projects) + 339 (loose) = 577

Deferred + open questions

• Deploy to ceodash: not done. Needs Sondra's explicit "deploy it" + her OK that 577 private conversations go on the (Cloudflare Access-gated) dashboard. Process: run build_claude_projects.py FIRST, then ceo-dashboard\deploy.bat (refresh.py -> build_handoffs.py -> wrangler). Consider adding build_claude_projects.py to deploy.bat. Run /app-safety-check before publishing.
• CLAUDE.md router rule: drafted, awaiting her "add it." Adds a rule to the project CLAUDE.md telling future sessions to grep PROJECTS/Brain Migration/INDEX.md before saying a past conversation can't be found.
• SOPs shareable with others: she asked for this; SOP HTML is self-contained. Offer here.now publish or PDF export. Not yet done.
• Live "find that chat" test: offered to prove retrieval; she hasn't run it.
• Deferred: topic tagging in the catalog over-tags "Business/Strategy" — refine later.

Pick up here

Once Sondra says "deploy it": run python ceo-dashboard\build\build_claude_projects.py then ceo-dashboard\deploy.bat to publish; then add the CLAUDE.md router rule and make the SOP shareable. Never restyle claude-desktop-projects.html — it must stay identical to MASTER-INDEX.