rock@ai-lab ~ % whoami --verbose
handle rockcli
role AI Agent Architect & Hands-on Practitioner
site https://rockcli.github.io
focus Multi-model harnesses · Agent skills & ops · Memory/privacy kernels · Observability
stack Python · Swift (macOS) · TypeScript / Node.js · Shell
- Multi-Agent Harnesses & Governance — Unbiased multi-model blind review (
Codex+Claude Opus+DeepSeek), cross-host session handoff, fail-closed privacy kernels, and grant-scoped memory engines. - Domain Agentic Pipelines — Knowledge-grounded strategy research loops (
Generate → Backtest → Diagnose → Iterate) paired with statistical anti-overfitting gates (PBO/CSCV,WFA,Embargo). - AI Observability & Native macOS Tooling — Quota & burn-rate telemetry, CDP HUDs & spatial agent workspaces, Ghostty TUI automation, and zero-dependency menu bar utilities.
- agent-ops-skills — 16 battle-tested Agent Skills for parallel multi-model blind review (
Codex/Claude Opus/DeepSeek), cross-host session handoff (codex://,dsh, Ghostty TUI), and safety-first developer ops. - soul-memory-kernel — Zero-dependency Python kernel for AI agent soul identity, fail-closed privacy gating, grant-scoped SQLite memory projection, and budgeted extraction.
- agentic-quant-lab — Knowledge-grounded agentic quant research pipeline (
RAG + Generate/Backtest/Diagnose/Iterate) and anti-overfitting validation (PBO/CSCV,WFA,Embargo). - antigravity-office — Lightweight, zero-dependency 2.5D Pixel Agent Office & CDP HUD for Google Antigravity on macOS, unifying live
Claude Code,Codex,OpenCode,DSH, andHowManyUsagetelemetry. - howmanyusage — Native macOS menu bar monitor for
Codex,Claude,Antigravity,DeepSeek, andColabquotas, burn-rate forecasts & model speed. - TypeMate — Zero-dependency native macOS menu bar writing assistant & instant double-
⌘Ctranslator powered byCGEvent+ DeepSeek. - ghostty-cc-bilingual — Zero-dependency Ghostty bilingual side pane for Claude Code on macOS with read-after-write JSONL verification, structural bold/code preservation, and local Ollama telemetry.
Design bias: zero unnecessary dependencies, fail-closed boundaries, deterministic verification around probabilistic models, and local-first telemetry.
