# 2026-04-04 — Saturday

## Summary
Claude Code local setup day. Assaf connected to the server via SSH and ran Claude Code for the first time.

## Key Decisions & Events

### Claude Code Setup (morning)
- Built `CLAUDE.md` — native boot file for Claude Code, loads Kitt persona + all memory files
- Built `.claude/settings.json` — model defaults and tool permissions
- Exported 11,832 Discord messages across 57 channels to `discord-history/` (4.1 MB)
- Updated `CLAUDE.md` to reference `discord-history/projects/mission.md` for Mission Mode context
- Assaf tested it via SSH — worked, Claude Code found CLAUDE.md and offered global install
- Kitt advised: say yes to global `~/.claude/CLAUDE.md`, pull from `work/migration/projects/kitt-ce-operations.md`
- Assaf asked if desktop Claude.ai would be Kitt too → Kitt: keep them separate (Kitt = server/SSH, regular Claude = desktop/Chat)

## Active Projects Status (unchanged)
- **Mission Mode:** Awaiting Assaf reaction to reference board v2
- **Olevia Brand Book:** Awaiting Assaf feedback on components.html
- **Cropster:** Waiting for Christina's status summary (due April 15)
- **WFP:** Lukas reviewing live page

## System Health
- All sessions healthy, no bloat
- No subagents running
- No new email

## HANDOVER
- **Discussed:** Claude Code local setup, SSH access, CLAUDE.md global install
- **Decided:** Kitt = server only, keep desktop Claude separate
- **Pending:** All active projects waiting on Assaf input (no blockers on Kitt's side)
- **Next:** Resume Monday or whenever Assaf picks up a project

## Spoken V2 — Late Session

### Context
Assaf wanted to access the Spoken V2 prototype at staging.curiousendeavor.com/spoken-v2/ but got a 404.

### Investigation
- Found that spoken-v2/index.html had been built (4 commits on main, last: `32feddf` — "unique content per topic") but was never on the staging branch.
- V1 prototype at spoken-language-v2/ was on both branches — no issue there.

### Fix
- Checked out spoken-v2/index.html from main into staging.
- Committed (`469c8db`) and pushed to staging.
- Verified 200 at staging URL.

### Status
- **staging.curiousendeavor.com/spoken-v2/** — live ✓
- **spoken-v2 still only has V1-era build** — Anton's review items (real audio, mobile, dead buttons, missing session steps) not yet addressed.

### Anton's Review of Spoken V2 (2026-04-03)
Full review at: work/active-projects/spoken/ANTON-REVIEW.md

**PRD — what's strong:**
- Philosophy-as-product is genuine (Krashen's input hypothesis, immersion-based pedagogy)
- POR as atomic unit is the best idea in the doc
- Language Braiding well-articulated
- 13 features internally coherent
- Build sequence realistic

**PRD — what's wrong:**
1. No opinion on what a session *feels like* — pacing, duration, emotional arc unspecified
2. The Dial is conceptually strong but functionally vague (user-controlled? automatic? what does it drive?)
3. 800-word vocab has contradictory sourcing logic (per-topic vs persistent)
4. "No stage labels" principle contradicts the prototype which shows labels
5. Grammar Sense feature is content marketing, not product — needs a stronger mechanic
6. Open Questions (§11) are still open — prototype was built before answering them
7. No error states, edge cases, or empty states
8. API key exposed in plain text in §9

**PRD — underspecified:** audio/TTS integration, recording storage (localStorage won't scale), joke/pun sourcing, Speech Work timeline  
**PRD — overspecified:** "5 argument-angles" is arbitrary, data model too early

**Prototype execution — confirmed Kitt's 6 issues + added 11 more:**
1. Z2 canvas cards overlap on <1200px viewports
2. Session card overlay too transparent (82% opacity, needs 95%+)
3. Vocab panel width hardcoded, overlaps session card on narrow screens
4. Stage label too small (10px)
5. Z1→Z2 has no real zoom transition (just opacity crossfade)
6. The Dial is tiny (110px×5px) — signature metaphor occupies less space than avatar
7. **No audio at all** — play button does nothing, mic fakes recording. Critical for a listening-first product.
8. Braided text hardcoded to one topic — every topic routes to street food session
9. SpoKen logo looks like a loading glitch
10. Mobile broken — vocab panel hidden, session card jumps at 600px breakpoint
11. Custom topic input button has no onclick handler
12. Breadcrumb pills fragile, clicking topic doesn't close session card
13. "Hear yourself back" and "Speech Work" cards are dead (no onclick)
14. No etymology, no idiom layer, no joke — 3 most distinctive features missing
15. "Hear yourself back" shows fake improvement waveform (contradicts "no grading" philosophy)
16. Portuguese color (#C0392B) shared with mic, waveform, Dial, stage labels — language competes with UI chrome
17. No favicon, no meta, no OG tags

**Anton's priority fix list (top 3):**
1. Add real audio (even one ElevenLabs clip transforms demo from wireframe to product)
2. Fix session card overlay opacity
3. Add real zoom transition for Z1→Z2