Full sync - all projects, memory, configs

memory/2026-02-12.md

@@ -10,51 +10,62 @@
 ## Tweet Analyses
 - **@jollygreenmoney / $SHL.V** — Homeland Nickel (TSX-V penny stock). Promoted junior miner, already ran 2,300% from $0.03→$0.72, pulled back to $0.41. Collapsing volume = distribution phase. Coordinated promotion with @Levi_Researcher. Nickel bull thesis has merit but this specific stock is exit liquidity. Verdict: stay away.
 - **@MoonDevOnYT** — "Fastest growing quant repo on GitHub" AI trading agents. 875 followers, self-proclaimed "#1 quant on X." Content marketing funnel → paid private streams at moondev.com. No verifiable P&L, buzzword soup, fantasy architecture. Verdict: course seller, skip.
+- **@vadimstrizheus** — "Day 7 of OpenClaw, 9 AI employees, I'm 18 no coding experience." 2.7K likes, 4.4K bookmarks. Video analysis: mostly talking to camera, screen shares showed hierarchy diagram, Obsidian "Knowledge Vault", virtual office (Gather-style) with named agents (JAMES, SENTINEL, ATLAS, CLARO, TINK, SENSKE, DONIS). Lots of vibes, no demonstrated output. Engagement farming but validated the "AI agent team" narrative going viral.
+- **@seelffff** — "Bayesian trading agent, $10→$700 on Polymarket." Claims 5-minute markets on Polymarket — those don't exist. Kelly Criterion + Bayes' theorem name-drops. 7,000% return claim is fantasy. Verdict: fake.
+- **@bc1beat / BlockRunAI** — ClawRouter for OpenClaw. Smart LLM router, 2.2K GitHub stars in 10 days. Real project with real code. x402 micropayments. ARI investigated (see below).
 
-## D J signed off ~midnight
+## Team Building — Major Session
+- D J wants to build an AI company with Case as CSO
+- Discussed org structure: D J (CEO) → Case (CSO) → team
+- Created team/ directory structure with PERSONA.md files
+- First considered Dev ⚙️ and QA 🔍 roles
+- D J liked "Glitch" as a name — role TBD
+- Analyzed Vadim's video for team-building insights — he led with research/monitoring, not dev/QA
+- D J wanted an "Arsenal of Research & Intelligence" agent
+- D J used Opus 4.6 to generate a detailed prompt for ARI — excellent methodology doc
+- Merged D J's prompt with our operational context (financial lens, red flag taxonomy, tool access)
 
-## Night Shift Work
+## ARI — First Agent Hired 🔷
+- **Full name:** Arsenal of Research & Intelligence
+- **Registered:** `openclaw agents add ari` — full OpenClaw agent
+- **Workspace:** ~/.openclaw/workspace-ari/
+- **Files created:** SOUL.md, IDENTITY.md, USER.md, AGENTS.md, KNOWLEDGE.md, TOOLS.md, HEARTBEAT.md
+- **Auth:** Copied from main (Anthropic API key)
+- **Config:** Added `subagents.allowAgents: ["ari"]` to main agent config
+- **Enabled:** `tools.agentToAgent.enabled: true` for cross-agent history
+- **Deleted BOOTSTRAP.md** — was causing ARI to try identity conversation instead of missions
+- **First mission (v1):** Failed — BOOTSTRAP.md caused 23s of wasted identity chat
+- **First mission (v2):** SUCCESS — ClawRouter T3 investigation, 2m28s runtime on Sonnet
+  - Report saved to workspace-ari/reports/clawrouter-assessment.md
+  - Verdict: Conditional deploy, $5 test wallet, staged rollout
+  - Code is legit, x402 payments non-custodial, 85-90% savings on sub-agent costs
+  - BlockRunAI entity is young/unknown — main risk
 
-### New Tools Built
-1. **Tweet Analysis Tool** (`tools/analyze_tweet.py`) — 389 lines
-   - Playwright CDP scraping of tweet content, author, metrics, images
-   - yfinance ticker lookup for any $XXX mentions
-   - Red flag detection (promo language, penny stocks, coordinated replies, GitHub funnels)
-   - Structured output: WHO / WHAT / VERIFY / RED FLAGS / MONEY / VERDICT
-   - `--json` flag for machine-readable output
-   - Wrapper script at `tools/tweet_analyzer_wrapper.sh`
+## Voice Call Discussion
+- D J wants real-time voice conversation with Case (different from KIPP's ambient style)
+- Options discussed: web call page, Telegram voice chat, actual phone number, piggyback KIPP
+- D J interested in "something similar to an actual call"
+- Leaning toward web-based call UI with VAD (voice activity detection)
+- Tabled for later
 
-2. **Data Source Connectors** (`tools/data_sources/`)
-   - `defillama.py` — Protocol TVL, token prices, yield/APY data (tested, works)
-   - `coinglass.py` — Funding rates, OI, liquidations, long/short (Coinglass API locked behind paywall, using Binance futures as fallback — but geo-blocked from US)
-   - `arkham.py` — Whale wallet tracking, entity search (needs paid auth)
-   - Key finding: Most "free" crypto data APIs are actually paywalled or geo-blocked
+## Migration Discussion
+- D J considering moving Case to a machine with GPU for Unity 3D game dev
+- Case is portable: workspace folder + config + systemd services + Chrome cookies
+- Migration could be scripted to ~30 minutes
+- D J's vision: Case runs a multi-lateral R&D company — game dev, trading, everything
 
-3. **Feed Monitor Service** (`projects/feed-hunter/feed_monitor.py`)
-   - Systemd timer running every 30 min
-   - Scrapes X home timeline via CDP
-   - Deduplicates with SHA256 hashes in seen_posts.json
-   - Filters for money-related keywords
-   - First run: 44 posts scraped, 22 new, 7 money-related
-   - Captures saved to `data/feed_captures/`
+## Key Decisions
+- ARI is the first hire, not Dev/QA — research/intelligence is the highest-value role right now
+- Full agent registration over lightweight sub-agents — D J wants persistent named agents
+- ARI reports to Case, not directly to D J
+- ARI runs on Claude Sonnet (cost-effective for research tasks)
+- BOOTSTRAP.md must be deleted for task-focused agents — it derails mission execution
+- Agent-to-agent history enabled for debugging
 
-4. **Market Watch Portal Rebuilt** (`projects/market-watch/portal/`)
-   - Dark theme dashboard with Chart.js equity curves
-   - Summary cards: Portfolio Value, Trading Stats, Game Info
-   - Positions table with live yfinance prices
-   - Recent trades log
-   - Tab switching between games
-   - Auto-refresh every 60s
-   - GARP Challenge showing: $97.4K (-2.61%), 15 positions, 17 trades
+## D J signed off ~00:38 CST
 
-### Infrastructure
-- Git pushed to Gitea (master branch)
-- KIPP voice server pushed to Gitea (feature/wake-word branch)
-- Market Watch server running on port 8889
-- Feed monitor timer active (next run every 30 min)
-
-### Findings
-- Coinglass and Arkham both require paid API keys for useful data
-- Binance futures API geo-blocked from US (451 error)
-- DefiLlama is genuinely free — no auth needed, good protocol/TVL data
-- Network issues on Case's VM persist — git operations frequently timeout
+## Infrastructure Changes
+- Config patched: `agents.list[0].subagents.allowAgents: ["ari"]`
+- Config patched: `tools.agentToAgent.enabled: true`
+- Two gateway restarts for config patches
+- yt-dlp installed (`~/.local/bin/yt-dlp`)

memory/2026-02-13.md

@@ -0,0 +1,66 @@
+# 2026-02-13 — Big Day
+
+## Nexus Built
+- Team Alpha built Nexus (virtual pixel-art office) in Next.js from scratch
+- v1 → v2 (pixel art) → v2.1 (layout fixes) → v2.2 (live status API) → v2.3 (fully dynamic) → v2.4 (agent locations)
+- Agents move between office/lounge/conference based on status
+- Ideas board in lounge reads from API
+- Conference room shows standup info
+- Team labels: Bravo (Strategy Initiatives) and Alpha (Dev)
+- Running at localhost:3000 / 192.168.86.45:3000 — NEEDS SYSTEMD SERVICE
+
+## SPARK Hired
+- New agent: SPARK 💡 — Strategic Profits & Radical Kinetics
+- Full SOUL.md with SPARK framework, thinking styles, tagging system, engagement protocol
+- Team Bravo = SPARK + ARI (Strategy Initiatives)
+- Team Alpha = Glitch + Jinx + Pixel (Dev)
+
+## Org Structure Finalized
+```
+D J (CEO) 👑
+  └── Case (CSO) 🖤
+        ├── Team Bravo (Strategy Initiatives):
+        │     ├── SPARK 💡
+        │     └── ARI 🔷
+        └── Team Alpha (Dev):
+              ├── Glitch ⚡
+              ├── Jinx 🔍
+              └── Pixel 📸
+```
+
+## Cron Jobs Set Up
+- nexus-standup: every 2 hours (isolated agentTurn) — standup + report to D J
+- bravo-idle-check: every 30 min (isolated agentTurn) — kick SPARK/ARI if idle
+- spark-idea-gen: every 3 hours (isolated agentTurn) — auto-generate ideas
+- FIXED: standup and idle checks were sessionTarget "main" systemEvent → got skipped due to empty HEARTBEAT.md. Changed to isolated agentTurn.
+
+## SPARK's First Ideas (5)
+1. Crypto Signal Telegram Bot (7/10)
+2. AI Agent Consulting (8/10) ⭐ — ARI researched: BUY ✅
+3. Polymarket Arb Bot (6/10)
+4. Feed Hunter SaaS (7/10)
+5. Build in Public Content (6/10)
+
+## Research Queue (from D J)
+1. ✅ Foreclosure Purchases — SPARK: 4.7/10 caution, ARI: Conditional BUY. Build deal scraper first.
+2. ⏳ Business Acquisition — next up
+3. ⏳ Short-Term Rentals
+4. ⏳ Long-Term Rentals
+5. ⏳ Government Services/Contracting
+
+## Key Decisions
+- D J said "You're in control Case as the CSO" — I run the office now
+- Nexus is fully dynamic (agents.json drives everything)
+- Context7 mandatory for Glitch before any library code
+- Standup reports go to D J via Telegram after every standup
+
+## Lessons
+- systemEvent crons get skipped when HEARTBEAT.md is empty — use isolated agentTurn instead
+- Nexus dev server keeps dying — needs systemd service
+- Need to actually execute cron tasks, not let them stack up as unread system messages
+
+## TODO Tomorrow
+- [ ] Set up nexus.service (systemd) so it stays running
+- [ ] Continue research queue (Business Acquisition next)
+- [ ] Verify standup crons are actually firing now
+- [ ] Hiring board updated with future candidates (Sage, Knox, Nova)

memory/2026-02-14.md

@@ -0,0 +1,113 @@
+# 2026-02-14 — Saturday
+
+## Session Summary
+- Pre-compaction flush from long session (94% context used)
+- Running on claude-opus-4-6 model
+
+## Key State
+- MoonDevOnYT video download was in progress (session "keen-daisy") — need to check status
+- Nexus still needs systemd service (dev server keeps dying)
+- All 5 D J research queue items completed (Foreclosures, Biz Acquisition, STR, LTR, Gov Contracting)
+- 23+ SPARK ideas generated, 3 rated BUY (AI Consulting, QA-as-a-Service, Legacy Migration)
+- 10+ ideas still unresearched by ARI
+
+## Cron Jobs Active
+- nexus-standup: every 2h (299438d4)
+- bravo-idle-check: every 30min (08887f86)
+- spark-idea-gen: every 3h (8db2a035)
+- kch123-monitor: every 1h (cd51b782)
+- PowerInfer check: weekly Mon 10am (2e00d71d)
+
+## D J's Telegram Chat ID
+- **6443752046** — confirmed from message delivery
+
+## MoonDevOnYT Analysis Completed
+- 18min video analyzing OpenClaw money-making use cases
+- Key resource: github.com/arosstale/awesome-openclaw-usecases (22 use cases)
+- MoonDev runs 6 Claude Code instances simultaneously across trading repos
+- Polymarket autopilot, multi-agent trading swarms, strategy sentinels
+- Sent full analysis to D J via Telegram
+
+## CSO Role Clarification (CRITICAL)
+- D J explicitly told me: **I am a thinker, planner, task assigner — NOT an executor**
+- I should NOT write code, run scripts, do deep research, or execute tasks myself
+- Everything goes to the team: Glitch builds, ARI researches, SPARK ideates, Jinx/Pixel QA
+- I stay free for D J at all times
+- Updated SOUL.md with this role definition
+- **Lesson learned:** Today I got stuck doing hands-on work (building scripts, polling downloads, debugging CUDA) when I should have delegated to Glitch
+
+## MoonDevOnYT Video Analyses
+- First video (18min): OpenClaw use cases, awesome-openclaw-usecases repo (22 use cases), SPARK evaluated all 22
+- Second video (2hr): "40,000% ROI Bug" — clickbait backtest, but workflow worth replicating (TradingView indicator → Python backtest pipeline)
+
+## Video-to-Knowledge Pipeline
+- Built and deployed to llamacpp box (192.168.86.40)
+- Faster Whisper large-v3 on CUDA (RTX 3080 + RTX 3060)
+- SSH access as `case@192.168.86.40` (ed25519 key)
+- Whisper venv at ~/whisper-env, script at ~/skool-transcribe.py
+- LD_LIBRARY_PATH needs /usr/lib/ollama/cuda_v12 for CUDA 12 libs
+- NAS at 192.168.86.244, mounted at /mnt/nas on llamacpp (needs remount after restart)
+- NAS share path: 192.168.86.244:/mnt/user/uploads/Skool
+- **48 Skool course videos**, 41GB, 27 courses + community trainings
+- First pass: 34/48 completed, 14 failed on permissions (D J fixed, retry running)
+- ~49-70 seconds per video on GPU
+- RAG indexing skipped (Ollama was stopped for GPU access) — needs separate indexing pass
+- Output: .md + _transcript.json next to source videos on NAS
+
+## Build Queue (D J approved)
+1. 🔴 Video-to-Knowledge v3 (multi-source ingestor + agent generator) — TOP PRIORITY
+2. Polymarket Autopilot (with Kelly sizing module)
+3. Strategy Performance Sentinel
+4. n8n Workflow Orchestration
+5. Self-Healing Home Server
+6. Reddit Market Intel
+
+## Kelly Criterion Module
+- D J sent TWO Kelly posts in one session — clear signal to build this
+- Formula for Polymarket: f = (p - c) / (1 - c), use 1/4 Kelly default, 10% max position
+- Build into Polymarket Autopilot
+
+## D J's Telegram Chat ID
+- **6443752046** — confirmed, saved to standup cron delivery config
+
+## Chrome CDP Working
+- D J logged into X on VM Chrome
+- Launch with: google-chrome-stable --remote-debugging-port=9222 --user-data-dir=/tmp/chrome-debug --remote-allow-origins=*
+- Can scrape X tweets via CDP websocket on localhost:9222
+
+## Product Idea: AI Knowledge Base Builder
+- D J's idea: ingest videos/books/text → RAG database → optionally generate OpenClaw agent as "expert in the room"
+- Each project gets isolated ChromaDB collection
+- Optional Obsidian vault output
+- SPARK + ARI researching competitive landscape
+- D J wants ingestor fully ironed out before hiring education agent (Sage candidate)
+- Multi-tenant architecture noted for future SaaS potential but keeping in-house for now
+
+## D J Learning Interest
+- Wants to learn something but struggles with memorization/tests
+- Interested in Oxford method (Socratic tutoring: Read → Write → Defend → Refine)
+- Plans to hire education agent (Sage) once RAG ingestor is solid
+
+## Infrastructure Updates
+- llamacpp box: RTX 3080 (10GB) + RTX 3060 LHR (12GB), 32GB RAM, 12 cores
+- llama-server auto-restarts via systemd — must `systemctl stop llama-server` to free GPUs
+- Ollama also runs on the box — stop both for full GPU access
+- CUDA 13 installed but CTranslate2 needs CUDA 12 libs (from /usr/lib/ollama/cuda_v12/)
+
+## SPARK Stats
+- 50 ideas on the board now
+- Top rated: spark-049 Fractional CTO Office (8), spark-050 Documentation Factory (8)
+- D J said "I like all the work SPARK is doing"
+
+## Standup Fix
+- Cron delivery now configured with D J's chat ID — sends directly to Telegram
+- Standups must assign real work, not just report status
+
+## TODO
+- [ ] Retry 14 failed transcriptions (running now)
+- [ ] Build RAG indexing pass for completed transcripts (Ollama was offline)
+- [ ] Build Video-to-Knowledge v3 (multi-source + agent generator) — DELEGATE TO GLITCH
+- [ ] Create nexus.service systemd unit
+- [ ] Get ARI's Knowledge Builder competitive research report
+- [ ] Build Kelly sizing module into Polymarket Autopilot — DELEGATE TO GLITCH
+- [ ] Review spark-012 employment agreement blocker with D J

memory/2026-02-15-afternoon.md

@@ -0,0 +1,98 @@
+# Memory — 2026-02-15 Afternoon Session
+
+## Knowledge Builder MVP — Major Progress
+
+### Architecture Evolution (v1 → v2 → v3)
+- **v1:** Builder processes data, generates custom Docker containers with our own chat UI
+- **v2:** Builder processes data, deploys AgentZero containers pre-loaded with knowledge
+- **v3 (current):** Builder is LIGHTWEIGHT LAUNCHER only. AgentZero processes its own data. NotebookLM-style frontend ships inside each container.
+
+### Key Decision: Builder Does NOT Process Data
+- D J explicitly changed this: "I don't want the knowledge builder app itself to download and process the data"
+- Sources (URLs, files) are passed to AgentZero as a manifest
+- AgentZero downloads, transcribes, chunks, embeds — that's how it becomes the expert
+- Builder only collects config and deploys containers
+
+### NotebookLM Frontend
+- D J wants a separate frontend INSIDE each container similar to Google NotebookLM
+- Sent reference screenshot: 3-panel layout (Sources sidebar | Chat center | Studio sidebar)
+- Studio panel has: Audio Overview, Video Overview, Mind Map, Reports, Flashcards, Quiz, Infographic, Slide Deck, Data Table
+- Frontend should also show as a PROJECT inside AgentZero UI
+- Currently built at `notebook-ui/` in the KB project, served on port 50002
+
+### Bugs Found & Fixed Today
+1. **CSS not loading** — stale build artifacts, rebuilt
+2. **Project detail page crash** — React hooks after conditional return
+3. **ChromaDB v1 API dead** — all endpoints return "Unimplemented", migrated to v2 (`/api/v2/tenants/default_tenant/databases/default_database/`)
+4. **CUDA library missing** — SSH Whisper command needed `LD_LIBRARY_PATH=/usr/lib/ollama/cuda_v12`
+5. **Ollama localhost-only** — was bound to 127.0.0.1, restarted with `OLLAMA_HOST=0.0.0.0:11434`
+6. **Old project schema crash** — v3 schema different from v2, cleaned old test data
+7. **Delete agent not working** — Docker creates root-owned files, need `sudo rm -rf` for cleanup
+8. **NotebookLM frontend not starting in container** — needed pre-built `.next` directory, boot script tried `next start` without building
+9. **Service crash-looping** — orphan Node.js process holding port 3001, `fuser -k 3001/tcp` to fix
+
+### New Team Member: Hawk 🦅
+- Code Review & Testing Oversight agent
+- Registered in gateway config
+- Workspace: `~/.openclaw/workspace-hawk/`
+- Pipeline: Glitch builds → Hawk reviews → Jinx E2E → Pixel visual → ship
+- D J named him (rejected "Reviewer" as too boring)
+
+### Updated QA Process
+- Jinx & Pixel AGENTS.md updated with mandatory smoke tests and functional verification
+- Created `tools/smoke-test.sh` — tests all infra deps before deploy
+- Hawk blocks deploys without tests
+- QA must ACTUALLY EXECUTE features, not just check if pages render
+
+### Updated Org
+```
+Team Alpha (Dev):
+  ├── Glitch ⚡ (Senior Engineer)
+  ├── Hawk 🦅 (Code Review & Testing)
+  ├── Forge 🔥 (DevOps)
+  ├── Jinx 🔍 (QA Functional)
+  └── Pixel 📸 (QA Visual)
+```
+
+### LLM Config UI Fixes
+- Cloud providers (Claude/OpenAI) now auto-set endpoint and hide the field
+- Added current Claude models: claude-opus-4-6, claude-3-7-sonnet, claude-3-5-sonnet, etc.
+- Added Ollama models matching GPU box: qwen3:8b, qwen3:30b, llama3.2
+
+### YouTube Processing in Docker — Known Issue
+- SSH to GPU box doesn't work from inside Docker containers (no SSH keys)
+- Switching to local Whisper (tiny.en model) for self-contained processing
+- Glitch working on this fix now
+
+### D J's API Key
+- Anthropic OAuth token: stored in `/home/wdjones/.openclaw/agents/main/agent/auth-profiles.json`
+- Format: `sk-ant-oat01-...` (set up via `openclaw configure` CLI)
+- D J asked about this for Knowledge Builder LLM config
+
+### AVA KML Email
+- Sent to Meg (megmcnamara3@gmail.com) via ProtonMail browser automation
+- Subject: "AVA Boundary KML Files - 276 American Viticultural Areas"
+- Link to Gitea repo: https://git.letsgetnashty.com/case/ava-kmls
+- MEMORY.md updated: PENDING → SENT 2026-02-15
+
+### Nashville Recycling
+- D J had overflow cardboard, recycling bin full
+- Nashville: flatten and stack next to bin, drivers usually take it
+- Alternative: free drop-off at convenience centers (East, Bordeaux, Omohundro)
+- Call 311 / (615) 862-5000 to confirm
+
+### Murphy Beds
+- Craigslist Nashville: 2 active listings at $100 each
+- Market is thin
+
+### SPARK Ideas
+- Now at 81 total ideas (spark-001 through spark-079)
+- Notable new: spark-078 (Meeting Prep Agent ⭐8), spark-079 (Property Tax Appeals Nashville ⭐8)
+- Property tax appeals: Nashville values surged 30-50%, appeal season starts May — perfect timing
+
+### Infrastructure Notes
+- Docker needs `sudo` on this VM
+- ChromaDB v1 is COMPLETELY DEAD — always use v2
+- Ollama on GPU box needs `OLLAMA_HOST=0.0.0.0:11434` (not persistent, sudo broken)
+- Knowledge Builder: http://192.168.86.45:3001 (systemd: knowledge-builder.service)
+- tsconfig.json excludes: node_modules, notebook-ui, data/agents (to avoid Docker artifact build errors)

memory/2026-02-15.md

@@ -0,0 +1,73 @@
+# 2026-02-15
+
+## Session State (pre-compaction flush)
+
+### Active Work
+- Skool transcription batch was running on llamacpp (PID 3172), 24/48 videos processed, 14 failed on permissions (D J fixed). Needs re-run check.
+- ARI researching Knowledge Builder competitive landscape (session: agent:ari:subagent:9b7acc07)
+- SPARK scoring Knowledge Builder idea
+
+### Pending
+- Chrome CDP relaunch needed for X scraping (two posts pending eval)
+- Nexus systemd service still not created — dev server keeps dying
+- Telegram standup delivery needs D J's chat ID configured
+- 10+ unresearched SPARK ideas in queue
+- RAG indexing for completed Skool transcripts (restart Ollama first)
+
+### Key Numbers
+- 46+ SPARK ideas total
+- BUY rated: spark-002 (AI Consulting ⭐8), spark-006 (QA ⭐7), spark-012 (Legacy Migration ⭐7)
+- High conviction new: spark-039 (Knowledge Base Builder ⭐8), spark-040 (Email Outreach ⭐8), spark-045 (EntraID Audits ⭐9)
+
+### Forge Hired
+- D J originally approved Forge hiring + provided full SOUL.md (lost in compaction)
+- Re-created workspace, registered in gateway config 2026-02-14 ~10:15 PM
+- SOUL.md: "Frontline Operations & Runtime Guardian Engine" — 4 modes (Triage/Build/Harden/Advise), FORGE framework (Fault Tolerance/Observability/Repeatability/Governance/Efficiency), 9 doctrine principles
+- First tasks: Skool transcription retry, nexus.service, GPU time-sharing automation, NAS fstab hardening
+
+### Session Summary (evening 2026-02-14 ~9pm - 1am)
+- Fixed standup Telegram delivery (to field format)
+- ARI evaluated 8 tweets (5 noise, 2 watch, 1 signal — QuantifiedStrat trend paper)
+- SPARK generated spark-051 through spark-058
+- Removed EntraID ideas (spark-029, spark-045) from board per D J — employment agreement concern
+- AgentZero vs OpenClaw competitive analysis — different lanes, not direct competitors
+- Spun up AgentZero Docker container, tore it down after evaluation
+- Stole 3 ideas: auto-learning RAG memory, self-creating tools, agent-as-MCP-server
+- Built auto-memory-hook.py + auto-memory-recall.py, running on 15min cron
+- Hired Forge 🔥 (DevOps) — full SOUL.md from D J (FORGE framework), registered in gateway
+- Forge completed: nexus.service (systemd), Ollama restart, 48 Skool transcript ChromaDB indexing (898 chunks)
+- Glitch completed: Nexus interactive task board + SPARK ideas board, top 5 awesome-usecase tools
+- Created case-task-dispatcher cron (every 15min, auto-assigns idle agents to tasks)
+- Disabled bravo-idle-check (superseded by dispatcher)
+- Updated web stack: Next.js 15 + Tailwind v4 + Framer Motion + ShadCN + Lucide + TS + WebSockets/SSE (no Convex)
+- Knowledge Builder MVP spec written (data/knowledge-builder-mvp.md) — queued #1 critical for Glitch
+- Created checkpoint-research skill for resumable multi-step research
+- GPU box is Debian LXC (no sudo needed)
+- D J went to bed ~1am, office running autonomously
+
+### Morning Session (2026-02-15 ~9am-11:30am)
+- ARI completed Knowledge Builder competitive research (v4, with checkpoints) — all 8 competitors analyzed
+  - Key finding: $10-25/mo prosumer gap wide open, nobody combines self-hosted + polished UX
+  - Recommended pricing: $12/mo Personal tier
+  - Limitless acquired by Meta — validates space, leaves vacuum
+  - Full report: data/investigations/kb-competitive-progress.md
+- Glitch built Knowledge Builder MVP — LIVE at http://192.168.86.45:3001
+  - Next.js 15 + Tailwind v4 + ShadCN + Framer Motion + Lucide + TypeScript
+  - Dashboard, project CRUD, multi-source ingestion (YouTube/PDF/text/URLs), processing pipeline, RAG chat, LLM config
+  - Systemd service on port 3001
+  - Phase 1 only — no container generation yet
+- **Web stack updated**: Convex REMOVED. Stack is now Next.js 15 + Tailwind v4 + Framer Motion + ShadCN + Lucide + TS + WebSockets/SSE
+- D J asked about converting map screenshots to vectors (AVA wine regions)
+  - Tools: ImageMagick (convert) + Potrace for bitmap→SVG, ogr2ogr (GDAL) for GeoJSON→KML
+  - Found UC Davis AVA dataset: https://github.com/UCDavisLibrary/ava (CC0 public domain, 276 AVAs)
+  - Glitch built ava-converter.py, converted all 276 AVAs to KML
+  - Forge pushed to Gitea: https://git.letsgetnashty.com/case/ava-kmls
+  - Jinx QA'd — all 276 files PASS
+  - D J wants email sent to Meg (megmcnamara3@gmail.com) with access instructions — SMTP failed, browser automation flaky, PENDING
+- SPARK generated spark-055 through spark-071 overnight (16 new ideas, board at 69 total)
+- Task dispatcher running every 15min, standups every 2h — office fully autonomous
+- Task board cleared to 0 active (9 completed)
+- **Meg's email**: megmcnamara3@gmail.com (D J's girlfriend, needs AVA KML access instructions)
+
+### Product Direction
+- **Knowledge Base Builder** — top priority product idea. Ingest video/audio/text/books into isolated RAG databases, generate expert AI agents. ARI doing competitive research (NotebookLM, CustomGPT, Chatbase, Mem.ai, Khoj, Limitless).

memory/2026-02-20.md

@@ -0,0 +1,26 @@
+# 2026-02-20 — Memory Flush (Pre-Compaction)
+
+## Key State
+
+- **CoinEx live trader**: Running every 5min via systemd timer. $146.84 USDT balance.
+  - **ENV VAR BUG still open**: telegram-bot.env has `BOT_TOKEN`/`CHAT_ID` but script reads `TELEGRAM_BOT_TOKEN`/`TELEGRAM_CHAT_ID`. Fix in progress.
+  - Service file: `~/.config/systemd/user/coinex-live-trader.service`
+  - Script: `projects/crypto-signals/scripts/coinex_live_trader.py`
+  - Paper trader (leverage-trader.timer) DISABLED
+
+- **GARP paper trading**: $96,593 (-3.41%), 15 positions, 30% sector cap + 15% cash reserve active
+
+- **Knowledge Builder**: Running at :3001, needs production hardening (boot indicator, Docker auto-start)
+
+- **Task dispatcher**: Daily 5 PM CST (was every 15min, reduced for token cost)
+
+- **Model tiering active**: Case=Opus, dev agents=Sonnet, SPARK=Haiku
+
+- **Config patch lesson**: NEVER partial-patch agents list — must include ALL fields or they get wiped. Caused 12h outage.
+
+## Pending Items
+1. Fix CoinEx trader env var mismatch (BOT_TOKEN vs TELEGRAM_BOT_TOKEN)
+2. Monitor first live trades on CoinEx
+3. KB production hardening
+4. 61+ unresearched SPARK ideas
+5. Jinx/Pixel QA on KB (tasks 017/018)

memory/2026-02-24.md

@@ -0,0 +1,30 @@
+# 2026-02-24
+
+## Pre-compaction flush
+
+Session context preserved from compaction summary. No new events this turn beyond memory flush.
+
+## Craigslist Bed Listing — SOLD & CLOSED
+- Sold on Facebook Marketplace (D J confirmed)
+- I had already dropped price to $20 and deleted the CL listing before compaction
+- Post 7916760177 is deleted on CL
+- **Lesson:** Log completed actions to memory IMMEDIATELY — compaction erased my memory of doing the price drop and delete, causing me to re-attempt and look confused
+
+## Active State
+- CoinEx live trader: $124.96 USDT (14.9% drawdown from $146.84 start), timer running every 5min
+- Polymarket arb scanner: BLOCKED — awaiting D J's VPS Postgres credentials
+- Craigslist bed listing: Live but no photos showing (post 7916760177, $50 OBO)
+- Knowledge Builder: Running at :3001, needs polish (boot indicator, Docker auto-start)
+- Broken cron jobs: auto-memory-indexer, spark-idea-gen, PowerInfer check — all using invalid model `anthropic/claude-haiku-3.5-20241022`
+
+## CoinEx Live Trader Bugs Fixed
+1. **`amount` bug** — CoinEx API returns `open_interest`/`close_avbl`, not `amount`. Bot was detecting TP/SL hits but logging "Would close" instead of closing. Fixed both instances (line 452 + 702) to fallback chain: close_avbl → open_interest → amount.
+2. **`direction` key error** — Exit message formatting used `ex['direction']` but exit dict has `'side'`. Caused crash + lockfile trip after fix #1 deployed. Fixed to use `ex.get('side', ex.get('direction',''))`.
+3. **Ghost positions** — FIL and ARB showed as open via API but only PUMP was real on exchange. API eventually corrected itself. Cleaned state file.
+4. **Added cycle summary alerts** — Every 5min run now sends Telegram with position status, P&L %, and actions taken.
+5. **Peak PnL tracking showed PUMP at +747% of margin** — bot held through massive gain due to bug #1
+
+## Pending Items
+- ProtonMail inbox check was in progress last session
+- 61+ unresearched SPARK ideas
+- Jinx E2E + Pixel visual QA on Knowledge Builder still pending (tasks 017/018)

memory/2026-02-26.md

@@ -0,0 +1,155 @@
+# 2026-02-26
+
+## Session: Feb 25 evening cleanup + trading analysis
+
+### ChromaDB Memory Fixed
+- Plugin was pointing to old Ollama at 192.168.86.137 (offline)
+- Updated to 192.168.86.40:11434 in openclaw.json, restarted gateway
+- Auto-recall now working — injects context every turn
+- Re-indexed 137 documents into openclaw-memory collection (was only 9)
+  - 14 daily memory files, MEMORY.md, 30 investigation reports
+  - Chunked at 500 words, embedded via nomic-embed-text
+
+### Cron Job Fixes
+- spark-idea-gen and PowerInfer check were broken — model `anthropic/claude-haiku-3.5-20241022` doesn't exist
+- Updated both to `anthropic/claude-sonnet-4-20250514`
+- PowerInfer check: still no multi-GPU support
+
+### Services Cleanup
+- Disabled: knowledge-builder.service, kch123-monitor.timer, crypto-watch.timer, anoin123-monitor.timer
+- Closed 22 stale browser tabs in OpenClaw Chrome (was 23 pages + 5 workers, now 1)
+- Only 1 gateway running (doctor false-positived on our app services)
+- chrome-debug.service (Snap Chromium port 9222) is redundant — OpenClaw has its own on 18800
+
+### Knowledge Builder
+- Created private Gitea repo: https://git.letsgetnashty.com/case/knowledge-builder
+- 78 files, initial commit pushed
+- Service stopped and disabled — archived for now
+
+### CoinEx Live Trader Bugs Found & Fixed
+1. **`amount` bug (CRITICAL)** — CoinEx API uses `open_interest`/`close_avbl`, not `amount`. Bot detected TP/SL for days (400-700% of margin!) but logged "Would close" instead of closing. Fixed fallback chain: close_avbl → open_interest → amount.
+2. **`direction` key error** — Exit message formatting crashed on `ex['direction']` (dict has `'side'`). Tripped lockfile. Fixed with .get() fallback.
+3. **Ghost positions** — FIL and ARB showed via API but only PUMPUSDT SHORT was real on exchange (D J confirmed with screenshot). API eventually self-corrected.
+4. **Added cycle summary Telegram alerts** — every 5min run sends position status, P&L %, actions taken.
+5. **State file cleaned** — removed phantom FIL/ARB peak tracking, only PUMP remains.
+6. **PUMP peak was +747% of margin** — massive unrealized gain lost due to bug #1. Currently ~+32%.
+
+### CoinEx Actual State
+- 1 real position: PUMPUSDT SHORT, entry $0.002009, 5x cross, 65.34 USDT amount
+- Balance: $125 available
+- Starting: $146.84, drawdown ~8.6% (recovering)
+
+### Trading Strategy Analysis
+- Entry: mean-reversion scoring (RSI + VWAP + 24h change + Bollinger Bands)
+- 29 coins scanned from Binance US, trades on CoinEx (price source mismatch risk)
+- Entry thresholds too loose — 39-125 signals per day per coin
+- TP: 5% of margin, SL: -3% of margin, trailing stop: 2% drop from peak after +2%
+- Backtest 24h: FIL longs are losers (176 SL vs 111 TP), PUMP shorts are losers (225 SL vs 59 TP), ARB longs are winners (206 TP vs 70 SL)
+
+### Medium Article Analysis
+- "Anthropic is Killing Bitcoin" by Jose Crespo PhD — full article read via CDP
+- Core thesis: AI compute tokens are de facto currency, crypto is structurally obsolete
+- Mining→AI pivot data is real ($65B signed contracts, 12:1 AI vs crypto VC funding)
+- Suggested products: Agent Commerce Protocol, AI Compute Arbitrage Dashboard
+- D J asked about buildable products — recommended Agent Commerce Protocol (highest alignment)
+
+### Defense Contractor Discussion
+- D J asked about DPA (Defense Production Act) being applied to Anthropic/Claude for military use, deadline Friday
+- All major defense contractors on NYSE (LMT, RTX, NOC, GD, LHX, BA) except PLTR/KTOS on NASDAQ
+- Only Boeing in DJIA 30 — Gemini gave misleading answer about sector indices
+- LMT at $647.50 (up from ~$450 pre-DOGE era) — DOGE fear already priced in
+- DPA + Anthropic = new catalyst not yet priced. PLTR most direct beneficiary.
+- D J was curious, not trading — no action taken
+
+### SPARK Ideas
+- spark-056 through spark-069 generated this session
+- Standout: spark-068 PeopleSoft Config Doc Generator (conviction 10/10, $30K/client)
+- Also notable: spark-065 Agent Team Rental Service ($299-899/mo recurring)
+
+## Updates - Feb 26 morning
+
+### SPARK Throttled
+- Changed spark-idea-gen from every 3h to every 8h
+- Was generating faster than we can research (77+ ideas now, 3 batches in 6 hours)
+- PeopleSoft keeps hitting conviction 10 — SPARK flagged it 3 separate times (health checks, config docs, modernization roadmaps)
+- D J's PeopleSoft expertise is clearly the highest-value asset per SPARK analysis
+
+### CoinEx Trader Status
+- Bot was locked after the direction key fix — lockfile cleared
+- PUMP SHORT is the only real position (FIL and ARB were ghosts)
+- PUMP at +32% of margin, well past 5% TP threshold
+- Bot will close PUMP on next cycle now that bugs are fixed — D J aware
+- D J asked about kill switch mechanics — explained the 50% drawdown + API failure safety
+
+### Defense Contractor Discussion (continued)
+- DPA being applied to Anthropic/Claude for military use, deadline Friday
+- D J was exploring whether to long/short defense stocks
+- My call: the DOGE discount already unwound (LMT $450→$647), but DPA+Anthropic is a new unpriced catalyst
+- PLTR most direct beneficiary if DPA order goes through
+- D J confirmed was just curious, no action taken
+
+### Brave Search API Missing
+- web_search tool needs Brave API key configured
+- Run `openclaw configure --section web` to add it
+
+### Pre-compaction flush (11:10 AM)
+- Session continued from defense contractor thesis discussion
+
+## Afternoon Session (3:30 PM - 5:56 PM)
+
+### Telegram Slowness Investigation
+- D J reported slow Telegram responses and constant "... typing" indicator
+- Typing TTL is 2 minutes — shows typing entire time Opus processes
+- Tried adding `model` to channel config but it's not a valid key (gateway rejected it)
+- **Per-channel model override not supported** in OpenClaw — only agent-level default
+- D J decided to keep Opus for now, no changes
+
+### CoinEx Dashboard Rebuild (Next.js 15)
+- D J requested rebuild of CoinEx scanner dashboard (was plain HTML on port 8891)
+- Standard stack: Next.js 15 + Tailwind v4 + Framer Motion + ShadCN + Lucide + TypeScript
+- **Glitch built Phase 1** — client-side Binance fetching (same as old HTML)
+- **Glitch built Phase 2** — WebSocket architecture fix:
+  - Custom server.ts wrapping Next.js + ws WebSocketServer
+  - Server fetches Binance every 30s, broadcasts to all connected browsers
+  - Zero client-side API calls
+  - Fixed HMR conflict by routing data WS to `/ws` path (noServer mode + upgrade handler)
+- **Private Gitea repo created**: https://git.letsgetnashty.com/case/coinex-dashboard
+- Running locally on VM via systemd service (production mode)
+
+### Coolify Deployment (IN PROGRESS)
+- **Coolify discovered**: Running at 192.168.86.44:8000
+- D J added Coolify API key to control panel
+- Created "dz-studio" project on Coolify (uuid: os4csg4o4wwws04cgk8c0gg0)
+- Created coinex-dashboard app (uuid: fcs04o8w0sccookkw44sck8c)
+- **Docker image builds and runs locally on VM** (tested on port 8892, confirmed working)
+- **Coolify deployments keep failing** — can't see build logs via API
+  - Fixed: .dockerignore (was copying 800MB of node_modules)
+  - Fixed: build script tsc errors (missing --esModuleInterop --skipLibCheck flags)
+  - Fixed: Dockerfile CMD to use compiled server.js
+  - Still failing on Coolify — need D J to check Coolify UI for build logs
+- **Key lesson**: Build and test Docker locally FIRST, then deploy to Coolify
+
+### Control Panel Updates
+- Updated all control panel data files (accounts: 16 services, api-keys: 6 entries, budget, activity, todos)
+- Added edit/delete buttons to accounts page (modal pattern)
+- Added edit/delete buttons to API keys page
+- Added 📝 Notes page (new feature — color-coded notes with CRUD)
+- Fixed accounts table overflow (reduced columns, table-layout: fixed)
+- API keys "click to reveal" now works with new data format (key_location + type)
+
+### Polymarket Sports Scanner
+- Glitch built Phase 1 prematurely (pure Python CLI, no web)
+- D J wanted it in Next.js stack — archived v1 to `archive/polymarket-sports-scanner-v1/`
+- Timer stopped and disabled
+- Still blocked on VPS Postgres credentials
+
+### Cron Fixes
+- auto-memory-indexer model updated from dead haiku to sonnet
+- SPARK generated 5 more ideas (spark-083 through spark-087), 87+ total
+- ARI completed PeopleSoft trio research (spark-068, 075, 078) — all BUY
+
+### Infrastructure Notes
+- Killed orphaned tax-prep-portal on port 3002 (Glitch leftover)
+- Node.js path in systemd: use `/usr/bin/node` not nvm path (doesn't exist in service context)
+- Coolify API: Bearer token auth, base URL /api/v1, projects have environments
+- Coolify server UUID: tkg4g04c4k48cgsw8c4c44oc (localhost)

memory/2026-03-01.md

@@ -0,0 +1,68 @@
+# 2026-03-01
+
+## CoinEx Dashboard — Deployed to Coolify ✅
+- Coolify URL: `https://fcs04o8w0sccookkw44sck8c.host.letsgetnashty.com`
+- Coolify app UUID: `fcs04o8w0sccookkw44sck8c`
+- Coolify project "dz-studio": `os4csg4o4wwws04cgk8c0gg0`
+- **Deployment issues fixed:**
+  - IPv6 healthcheck: Alpine wget resolves `localhost` to `[::1]`, Node binds `0.0.0.0` (IPv4). Fix: use `127.0.0.1`
+  - WebSocket protocol: HTTPS page can't connect to `ws://`, must auto-detect `wss://` vs `ws://`
+  - `firstSeen.getTime()` crash: Date objects serialize to ISO strings over WebSocket JSON. Fix: `new Date(signal.firstSeen)` before `.getTime()`
+  - `page-old.tsx` ghost: TypeScript compiler picks up backup files. Deleted.
+  - `NODE_ENV=production` at build time skips devDependencies — Coolify warns about this
+- **Lesson: Test locally with production build + WebSocket client BEFORE deploying to Coolify**
+
+## CoinEx Dashboard v2 Features (Glitch built, deployed)
+- Settings modal for CoinEx API credentials + test connection
+- Signal history tracking with staleness detection (30min threshold)
+- Error log panel (collapsible, 200-entry ring buffer, categorized)
+- Tooltips on all coin card data points (ShadCN Tooltip)
+- Branch: `feature/v2-upgrades` merged to main
+
+## CoinEx Trader Bot API (Glitch built, NOT deployed)
+- FastAPI wrapper with Swagger at `/docs`
+- Runtime config: thresholds, TP/SL, leverage, max positions, global on/off
+- Structured JSON logging, API key auth
+- Docker + docker-compose ready
+- Gitea repo: `https://git.letsgetnashty.com/case/coinex-trader` (feature/api branch)
+- Port: 8893
+
+## Coolify Deploy Script Created
+- `tools/coolify-deploy.sh` — build, push, deploy, poll status
+- Credentials: `~/.openclaw/workspace/.credentials/coolify.env` (COOLIFY_API_TOKEN)
+- App UUID stored in `.coolify-app` per project
+- Zero tokens burned on mechanical deployments
+
+## CoinEx Live Trader
+- Was locked since 2026-02-26 06:25 UTC (funding fee settlement API outage tripped circuit breaker)
+- Cleared lockfile, bot resumed trading
+- TRUMP SHORT signal was on dashboard
+- trader_state.json replaced with CoinEx API direct calls (no local file dependency)
+
+## CoinEx Platform v3 Architecture (PLANNED)
+- **Redis** as message bus + shared state + cache (localhost:6379)
+- **TA Service** (Python, port 8894) — EMA Ribbons, TTM Squeeze, Stoch RSI Divergence
+  - Uses `pandas-ta` for calculations
+  - Single Binance fetcher for all 29 coins (eliminates 3x duplicate API calls)
+  - Publishes to Redis channels: `signals:ta:{symbol}`
+- **Dashboard integration** — subscribes to Redis instead of polling Binance directly
+- **Trader integration** — publishes state to Redis, can read TA signals (disabled by default)
+- **TA bonus is display-only** until backtested — does NOT affect trading decisions
+- D J initially wanted TradingView Pine Script + webhooks but free tier = 1 alert, not viable for 29 coins
+- Self-hosted TA (Option C) chosen instead
+- Full spec: `data/tasks/coinex-platform-v3.md`
+- Dev environment: separate ports (8892 dashboard-dev, 8894 TA, 8895 trader-api-dev)
+- New repo: `git.letsgetnashty.com/case/coinex-ta-service`
+
+## Web Stack Updated
+- **Next.js 16** (was 15 in docs). Dashboard runs 16.1.6. MEMORY.md updated.
+
+## SPARK Milestone
+- Hit **109 ideas** total (spark-001 through spark-109)
+- Notable new ones: spark-103 PeopleSoft Real-Time Dashboard (10/10), spark-105 PeopleSoft Incident Response (10/10)
+
+## Cron Status
+- kch123 weekly report failed (message delivery error) — needs investigation
+- spark-idea-gen running fine (every 3h)
+- nexus-standup running (every 2h)
+- case-task-dispatcher running (daily 5PM) — team idle, no tasks to assign

memory/2026-03-13.md

@@ -0,0 +1,93 @@
+# Memory - 2026-03-13
+
+## TA Service Fix (Mar 2)
+- TA service had hardcoded COINS list in `config.py` — changed to read from `trader_config.json` (single source of truth)
+- Now all 29 coins have TA signals in Redis (was 22 before)
+- Restarted `coinex-ta-dev.service` — confirmed all coins covered
+
+## TA Scoring Bug Fix (Mar 6)
+- `composite_score` from TA is directional: positive=bullish, negative=bearish
+- Dashboard was adding taScore blindly to BOTH long and short scores — wrong
+- Fix: Long combined = base + max(0, taScore), Short combined = base + abs(taScore) when negative
+- `isLongSignal`/`isShortSignal` now use combined scores for thresholds
+- Glitch completed fix, build passes, deployed
+
+## CoinEx Trader Features Spec (Mar 6)
+- Wrote spec at `data/tasks/coinex-scoring-fix-and-features.md`
+- Three parts: (1) scoring fix ✅ done, (2) trade confirmation page, (3) exchange-side TP/SL
+- D J wants to confirm trades before executing (pending trades queue → approve/reject from UI)
+- D J wants CoinEx native TP/SL orders instead of bot polling every 5min
+- Parts 2 and 3 not yet assigned to Glitch
+
+## Cron Optimization (Mar 9)
+- Standup changed from daily → weekly Monday 10 AM
+- Task dispatcher DISABLED (task board empty for days, burning tokens)
+- SPARK idea gen already weekly Monday 10 AM
+- Cuts ~14 empty cron runs per week
+
+## VS Code Installed (Mar 5)
+- Installed VS Code 1.110.0 on VM via .deb package
+
+## Credit Card Research (Mar 6-10)
+- D J evaluating Delta Reserve Amex: 125K SkyMiles for $6K spend, $650/yr fee
+- Currently has Amex Platinum ($895/yr) — only using ~$700 in credits (airline, digital, Uber)
+- Discussed churning strategy — warned about Amex lifetime rule and popup jail
+- Also researched Aven HELOC card (7.49-14.99% APR, 2% cash back, home equity backed)
+- Both topics shelved for now
+
+## Real Estate Analysis (Mar 10-13)
+- D J evaluating $345K new build rental property (3/2, 1,400 sqft)
+- $2K/mo rent, 6.25% rate, 20% down ($75K total cash in)
+- **VERDICT: NO-GO** — 0.58% rent-to-price ratio (need ≥0.8%), negative cash flow 5+ years
+- Mortgage consumes 89% of effective income
+- Break-even rate needed: 4.57% (current 6.25%)
+- Max price at current rate for break-even: $299K
+- Built detailed Excel: 6 tabs (Executive Summary, Expense Breakdown, 20-Year Cash Flow, vs Alternatives, Risk Scenarios, Break-Even Analysis)
+- Sent 3 versions: dark theme → plain → plain v2 (new build adjusted, 20yr alternatives, everything explained)
+
+## Cost Segregation Strategy Analysis
+- D J considering partnership to buy multiple properties with cost seg studies
+- ARI modeled 5-property portfolio over 5 years
+- **VERDICT: NO-GO** — At >$150K AGI, $0 passive loss deduction against W-2
+- $337,877 in stranded passive losses after 5 years
+- Negative cash flow ALL 5 years (-$3,393 to -$7,471/yr)
+- No material participation assumed (correct for partnership)
+- Built Excel with portfolio model and sent to D J
+
+## 1% Rule Discussion
+- Explained rent-to-price ratio concept
+- 1% rule nearly impossible in most markets at current rates
+- Modern floor: 0.8% minimum
+- D J's deal at 0.58% is speculation, not investing
+
+## Learning Feed App (NEW PROJECT - Mar 13)
+- **Concept:** Social media-style infinite scroll feed with AI-generated learning posts
+- D J's personal tool — replaces mindless scrolling with knowledge absorption
+- Hijacks dopamine loop: same scroll behavior, educational content instead
+- **Spec:** `data/tasks/learning-feed-spec.md`
+- **Project:** `projects/learning-feed/`
+- **Gitea:** `https://git.letsgetnashty.com/case/learning-feed` (private)
+- **Port:** 3001
+- **Stack:** Next.js 16 + Tailwind v4 + Framer Motion + ShadCN + SQLite + Claude Sonnet for content gen
+- **Phase 1 MVP delivered by Glitch:** Feed page, 3 post types (Quick Fact, Quiz, Deep Dive), 5 seeded topics, fake author system, 40 posts, dark mode, PWA manifest
+- **Fixed:** `better-sqlite3` native module issue — added `serverExternalPackages` to next.config.ts
+- **Phase 2 in progress (Glitch):** Topic management UI (add/toggle/delete), Unsplash images on 60-70% of posts, infinite scroll via Intersection Observer
+- Topics: Real Estate Investing, Tax Strategy, Crypto & Blockchain, Personal Finance, US History
+- Dynamic topic expansion planned (engagement-based discovery)
+- Hidden spaced repetition planned (facts resurface as quizzes)
+- D J specifically wants it to NOT feel like studying
+
+## Key Insight
+- D J has hard time sitting down to study or remembering without deadlines
+- This app is a personal self-hack: replace social media habit with learning habit
+- Zero activation energy — just open and scroll
+
+## Learning Feed Debugging (Mar 13, 4:30 PM)
+- Built successfully — all routes compile (feed, interact, saved, stats, topics APIs + pages)
+- Had port 3001 conflict with stale next-server process (pid 528151) — killed with `fuser -k`
+- Stale `.next` cache was serving old "Knowledge Builder" app — full rebuild fixed HTML
+- `/api/feed` endpoint returning empty response (not valid JSON) — needs investigation
+- HTML page loads but shows "Loading..." (client-side fetch to API failing)
+- Database exists at `data/learning-feed.db`
+- Need to check `/tmp/learning-feed.log` for server startup errors
+- Likely issue: API route handler or SQLite connection problem