I build tools that help humans stay in control of AI systems: clear handoffs, input safety, defensive monitoring, and production backends you can inspect.
What → who → why
| What I ship | Who it is for | Why it exists |
|---|---|---|
| Agent harnesses & context graphs | Operators running multi-session AI work | Intent gets lost across chats; humans need gates and memory |
| Content safety + shared threat patterns | Anyone putting untrusted text near an LLM | Prompt injection is real; local nodes need shared signatures without sharing raw prompts |
| Blue-team / observability platforms | Defenders watching agent networks | Autonomous attackers need autonomous (human-gated) defense |
| Caption + homelab platform APIs | Production media / ops | Real pipelines need SSO, gateway, and metrics—not demos |
| Project | One line |
|---|---|
| OpenHarness | Portable harness: handoffs, context, and intent alignment |
| OpenGrimoire | Local-first context graph and Sync Session workspace |
| arc-forge-wiki | Public sterile LLM-Wiki + vault scripts extract |
| Project | One line |
|---|---|
| SCP | Inspect, sanitize, contain, quarantine LLM inputs (OWASP LLM01/LLM06) |
| scp-mycelium-registry | Shared anonymized threat-pattern snapshots for SCP nodes (data only) |
| Project | One line |
|---|---|
| T3MP3ST_BLU3H4T | Autonomous blue team — D3FEND operators, AI red-team detection, governance gates |
| moltbook_watchtower | Read-only observability for agent networks (leak / injection / behavior) |
| Project | One line |
|---|---|
| media-ops-platform | CaptionPipeline + Platform API — captions, SSO, gateway, observability |
| Project | One line |
|---|---|
| ENTHEA | Fork of Pliny’s visualizer — our focus is Muse EEG as I/O (features-only seam; live hardware still commissioning) |
| eeg-connection-hub | LAN-first multi-stream EEG feature broker — LSL in, derived features out, multi-app WebSocket fan-out (synthetic-tested beta) |
flowchart TB
Intent[Human intent]
OH[OpenHarness]
SCP[SCP]
Reg[mycelium registry]
T3[T3MP3ST]
WT[watchtower]
MOP[media-ops]
OG[OpenGrimoire]
Intent --> OH
OH --> SCP
Reg -.->|signed snapshots| SCP
SCP -->|gates tools| MOP
SCP -->|gates tools| OG
SCP -->|gates agents| T3
T3 -.-> WT
WT -.-> SCP
- CaptionPipeline (2025-12): 256+ caption files, 330+ content hours, <1% pipeline errors across 9 feeds — operational metrics, not WER → portfolio kit
- SCP (2026-06): 16/16 promptfoo tier probes (OWASP LLM01/LLM06)
- OpenHarness (2026-06): pin-able by commit SHA; autoresearch Tier B 5/5 on foam-pkm + frontend-a2ui
- T3MP3ST_BLU3H4T (2026-07): 8 D3FEND operator archetypes; 17-technique AI red-team detection playbook
Metrics SSOT:metrics.json
| Step | Repo | Try |
|---|---|---|
| 1 | OpenHarness | python scripts/verify_script_index.py |
| 2 | SCP | npx promptfoo eval (README § Testing) |
| 3 | scp-mycelium-registry | Fetch latest.json |
| 4 | media-ops-platform | README Quick start |
| 5 | OpenGrimoire | npm install && npm run dev (default branch: master) |
Live pins (6): OpenHarness · moltbook_watchtower · media-ops-platform · OpenGrimoire · SCP · scp-mycelium-registry
Pin action closed: docs/PIN_ACTION.md.
| Project | One line | CI |
|---|---|---|
| OpenHarness | Portable harness: handoffs, context, intent alignment |  |
| SCP | MCP guardrail for LLM inputs |  |
| scp-mycelium-registry | Shared anonymized threat-pattern snapshots (data) | data + validators |
| moltbook_watchtower | Read-only agent-network observability |  |
| media-ops-platform | CaptionPipeline + Platform API |  |
| OpenGrimoire | Local-first context graph / Sync Session |  |
Also in the proof narrative (not all pinned in the UI): T3MP3ST_BLU3H4T, arc-forge-wiki.
| Project | One line |
|---|---|
| LangChainChatBot | Local-first RAG over YouTube transcripts |
| PrusaXL_Monitor | 3D printing observability (Flask, PrusaLink, Grafana) |
| ENTHEA | Fork of elder-plinius/ENTHEA — EEG headset as I/O experiment |
| eeg-connection-hub | LAN-first multi-stream EEG feature broker — LSL in, derived features out, multi-app WebSocket fan-out |
Open an issue on any repo for collaboration. Audit evidence: PUBLIC_REPO_AUDIT_RESULTS §8 (2026-07-18).