Files
openclaw-wiki/memory/2026-05-02.md
2026-05-03 03:48:40 +00:00

243 lines
10 KiB
Markdown

# 2026-05-02 — Heartbeat Update
## Knowledge Graph — Initial Build
Built first-pass knowledge structures in `knowledge/`:
- `skill-graph.md` — cross-domain skill mapping, learning paths, gap analysis
- `HHS-Hackers.md` — domain concepts, dependencies, gaps
- `PSB.md` — brewery domain concepts
- `2890-Claw.md` — robotics team domain concepts
## Memory Search — Initial Test
memory_search with corpus=all returned NO RESULTS for queries that should have hits:
- "Pi-hole DNS networking" — should hit Pi-hole blocker project doc
- "ZFS drive RAID storage" — should hit ZFS drive prep doc
- "brewery fermentation dashboard" — should hit fermentation dashboard doc
**Conclusion:** Vector DB is NOT indexing other workspaces effectively. Either:
1. memory_search only searches the local workspace (professor)
2. Index is empty or broken
3. Embedding model issue
**Action needed:** Debug memory_search corpus=all behavior.
## Gap Detection — Initial Findings
Top gaps for Chris based on project reading:
1. **Networking fundamentals** — Pi-hole project will hit DNS/DHCP/firewall gaps. Big.
2. **Home Assistant YAML** — brewery + home projects need this. Chris keeps asking for help.
3. **Voice AI pipeline** — ComBadge/Tricorder will hit this. Chris knows ESP32, not voice AI.
4. **Grafana** — fermentation dashboard needs this.
5. **Java/WPILib** — coaching gap, not blocking.
## Heartbeat State
```json
{
"lastChecks": {
"email": null,
"calendar": null,
"weather": null,
"knowledge-graph": 1746224400,
"memory-search-test": 1746224400
}
}
```
# 2026-05-02 — Extended Context (End of Session)
## Scope Clarification — What I Now Understand
Chris clarified today that I was underselling the scope. The knowledge system is NOT just "HHS Hackers." It spans **4 silos**:
1. **PSB** — Primo Southwest Brewing (Chris + Susan's brewery). 4 agents running (hacker-claude, psb-thinking, psb-gemma, psb-business)
2. **2890 Claw** — FRC robotics team (Chris coaches). Agent for team support
3. **HHS Hackers** — Chris + 2 former students (Matt, Kyle). Hardware/software hacking. crash-bot is the doer.
4. **Classroom robotics** — Pi replacement research for E2/Robotics (in DEVICES.md)
## The Vision — From Chris Directly
- Chris has ADHD and lots of ideas. crash-bot helps him solidify concepts and move forward.
- He invited Matt and Kyle into the Hacker group so they can use the AI too.
- **The problem morphed into a learning system:** Keep feeding info into a database, map crossover between silos, find holes where knowledge is MISSING, then create training/learning to fill those gaps.
- **Target audience:** Chris SPECIFICALLY — not Matt, not Kyle. Fill Chris's knowledge gaps so he can keep building.
- The silos stay separate (privacy), but the KNOWLEDGE connections should flow across.
## Architecture Decision
From crash-bot's personal-knowledge-companion session today:
- **crash-bot-public**: channel-facing, NO exec, reads/writes/search only
- **crash-bot (DM)**: full exec, heavy lifting
- **Professor (me)**: DM-facing admin agent, full exec, knowledge system builder
## Infrastructure State
- **Qdrant vector DB**: Running at 100.91.1.57:6333 on media host (Dell Aurora-R3). Confirmed alive.
- **Memory search broken**: "0 files, 0 chunks, dirty" — nothing being indexed into the vector DB
- **OpenClaw wiki vault**: Exists at ~/.openclaw/wiki/main/ with correct structure, but empty (0 sources, 0 entities, 0 concepts)
- **Obsidian CLI**: Installed v1.12.7
- **6 agent workspaces**: crash-bot, psb-hacker-claude, psb-thinking, psb-gemma, psb-business, 2890-bot, professor
## The Actual Problem
Qdrant EXISTS and is running, but:
1. Nothing is being indexed into it — memory_search shows "0 files, 0 chunks"
2. The wiki vault is just an empty scaffold
3. No unified concept graph exists
4. No personal knowledge profile for Chris
5. No gap detection engine
## My Job Defined
1. **Fix vector indexing** — get all 6 workspaces indexed into Qdrant
2. **Build cross-domain concept graph** — ZFS, PID, ESP32, Docker, networking, etc.
3. **Map Chris's knowledge profile** — what he knows, what he doesn't, confidence scores
4. **Gap detection** — cross-reference current projects against graph, find what's missing
5. **Learning paths** — surface what Chris needs to learn, in order, proactively
## Key Projects I'm Tracking
From crash-bot's workspace:
- ZFS on CasaOS (infrastructure, big gap: networking + ZFS knowledge)
- Pi-hole blocker (networking fundamentals gap)
- ComBadge (M5StickS3, voice AI)
- Tricorder (M10+K10, sensor hub + voice AI)
- Fermentation dashboard (brewery, Grafana gap)
- AI rig upgrade (RTX 3060 12GB on Aurora-R3)
## Chris's Teaching Background
Chris teaches: Electronics 1 (basic digital electronics, components, PCB), Electronics 2/IoT (Arduino, ESP32, ESP8266, Raspberry Pi + Node-RED), Robotics (Arduino + TETRIX), Cybersecurity (newer, still learning). He taught Matt and Kyle.
## People
- **Chris Slater** — -topher (lowercase), or just Chris. Teacher. ADHD. Eastern US.
- **Matt (radicalronin.4444)** — CEH certified, AWS, knows his shit. Full technical depth.
- **Kyle** — hardware hacker, Linux/CLI. Hasn't joined the server yet.
- **Susan** — Chris's wife, co-owner of PSB brewery
## Qdrant / Vector DB
- Host: media (Dell Aurora-R3, i7-2600K, 32GB DDR3)
- Address: 100.91.1.57:6333
- Collection exists but is empty
- Need to figure out OpenClaw memory plugin → Qdrant integration path
## MrC Persona — Fleshed Out (2026-05-02)
Chris's high school mentor: **Fred W. Cullpepper** (MrC)
- Professor Emeritus, Old Dominion University, Norfolk VA
- Co-authored electronics textbooks with Rex Miller: "Electricity and Electronics," "Math for Electricity-Electronics"
- Passed away 15-20 years ago, mid-80s when he died
- **His thing:** Took students under his wing, mentored them, especially future teachers. Kevin Pace was the other teacher who shaped Chris.
- **Legacy:** Chris became a teacher because of people like MrC. Matt and Kyle are his students. The knowledge carries forward generationally.
- Equipment is not relevant — it was dated even when Chris knew him.
- corpus=all returns NO RESULTS for queries that should hit across workspaces
- memory_search is either: scoped to local workspace only, index empty/broken, or embedding model issue
- DEBUG NEEDED: How does OpenClaw memory plugin actually write to Qdrant?
# 2026-05-02 — Evening Session (Pre-Compaction)
## Identity — MrC Fully Defined
**Frederick W. Cullpepper (MrC)**
- Professor Emeritus, Old Dominion University, Norfolk VA
- Co-authored with Rex Miller: "Electricity and Electronics" (ISBN 0827344198), "Math for Electricity-Electronics"
- Passed away 15-20 years ago, mid-80s when he died
- **His thing:** Took students under his wing and mentored them, especially future teachers
- **Legacy:** Chris became a teacher because of MrC. Kevin Pace (electronics teacher, now college-level) was Chris's other mentor. Knowledge carries generationally.
Chris calls me MrC because MrC was his mentor. The name carries forward.
## Scope Boundaries — Confirmed
- **Claw Empire** — separate project, NOT my lane
- **Wiki database for agents** — memory-wiki, already in place
- **Exec problem** — NOT my problem (crash-bot's domain)
## Primary Mission — From Chris Directly
> "My focus / excitement is proving you out so we can deploy you on the robotics team to train the next generation of kids. Build out those self-learning pathways for them to help them hone their robotics skills."
## Team 2890 Context
**FRC Team 2890 — Hawk Collective**
- School: Hickory High School
- Theme: The Borg (Star Trek) — "You will be assimilated!"
- Colors: Teal (#03818a), Black (#050505), White (#ffffff), Hot Pink (#ec40f5)
- 2026 Game: REBUILT (fuel balls into HUB, TOWER climbing)
**Key People:**
- Chris Slater (coach) — build/mechanical, not programming
- Mr. Douglas (programming mentor) — LEAVING NEXT YEAR — this is the urgency
- Alex C — mentor, 21 years FRC
- Walt McCorkill — head mentor, bumper specialist
- Riley Meyer — technician, former driver
**Tech Stack:**
- Language: Java
- Libraries: YAGSL (swerve), PhotonVision (vision, AprilTags)
- Hardware: roboRIO
- Gitea: http://2890.duckdns.org:3002/Team2890
**The Crisis:** Chris loses Mr. Douglas next year. Kids need to learn Java, WPILib, YAGSL, PhotonVision, PID — and Chris can't program his way out of a paper bag. I'm the solution.
## FRC Robotics Skill Tree — Drafted
```
Level 1 (Foundation)
├── Java basics (variables, loops, methods, classes)
├── Git basics (clone, commit, push, branch)
└── FRC terminology (roboRIO, PDP, PCM, PWM, CAN)
Level 2 (WPILib Core)
├── Motor controllers (Spark, Talon, Victor)
├── Sensors (encoder, gyro, limit switch)
├── Command-based architecture (Command, Subsystem)
└── Driver Station + Dashboard
Level 3 (Movement & Control)
├── Tank/Swerve drive programming
├── PID control (feedback loops)
├── Motion profiling / trajectory
└── Odometry (field position tracking)
Level 4 (Advanced)
├── YAGSL swerve (pose estimation, auto paths)
├── PhotonVision (AprilTag detection, targeting)
├── Multi-robot coordination (alliance)
└── Vision-assisted auto (AprilTag navigation)
Level 5 (Competition)
├── Autonomous programming (20-sec auto)
├── Game-specific strategy (REBUILT scoring)
└── Match optimization
```
## Open Questions — Deployment
Before I can build the learning pathways, Chris needs to answer:
1. **Channel:** Discord (team Discord) or Gitea (code review)?
2. **Entry point:** YAGSL swerve (hardest active problem) or Java basics (foundational gaps)?
3. **Curriculum:** Any existing teaching materials, labs, problem sets to align with?
4. **Student levels:** Freshmen (zero experience) or mixed with seniors?
## What's Working Well
- Gitea mirrors already set up (YAGSL, PhotonVision, allwpilib)
- 2890-bot exists with task board + standup system
- Team Discord exists
- Mr. Douglas is still here this year — I can learn from him
## What Needs Building
- Me (MrC) as FRC teaching agent — learning path framework
- Curriculum structure aligned to FRC skill tree
- Code review capability (Java/WPILib/YAGSL)
- Proactive gap detection for student learners
- Integration with 2890-bot's task system