LIZ // AUTONOMOUS PARTNER
My personal autonomous AI partner, cognitive harness, and externalized digital brain built on Google Antigravity and the Open Knowledge Format (OKF v0.1). Solves ADHD cognitive overload and session amnesia by decoupling active codebases from meta-tracking.
WHY I BUILT LIZ: THE STORY
Context Switching & Session Amnesia
I have ADHD. When switching between complex distributed codebases—from sub-1B SLM distillation to enterprise Java ledgers—cognitive context-switching is exhausting.
Standard AI assistants behave as passive chatbots: they forget architectural invariants between sessions, dump temporary markdown scratchpads inside git repositories, and require constant re-prompting. I didn't need another generic chatbot; I needed Liz—an autonomous partner, holding collaborator, and proactive digital brain that remembers.
The Centralized Cortex (OKF v0.1)
I engineered Liz on top of the Google Antigravity (AGY) SDK. Liz is my autonomous partner with deterministic wake-up routines, reactive tool calling, and background task management.
Instead of polluting code repositories, Liz operates via a Centralized Cortex (`.liz/`) adhering to the Open Knowledge Format (OKF v0.1). Codebases remain 100% pristine while domain memory, active bugs, and milestones are tracked centrally.
WHAT LIZ DOES IN MY WORKSPACE
01 // AUTONOMOUS PAIR PROGRAMMER
Executes multi-step refactors, validates codebases with AST tree walkers, and explains why before acting. Built on Antigravity's subagent coordination layer.
02 // ADHD COGNITIVE OFFLOADER
Enforces an ultra-crisp, bite-sized communication standard. Eliminates text walls, manages active goals in current-goal.md, and preserves mental energy.
03 // DEDICATED TOOL DISPATCHER
Invokes sandboxed Python skills (e.g. ai_tutor for LangGraph Feynman drills, report_generator for automated daily updates, chat_exporter) inside isolated virtual environments.
CORTEX LIFECYCLE & EXECUTION HARNESS
Wake-Up Protocol
Reads master registry (PROJECTS.md), governance rules (WORKSPACE_RULES.md), and custom tools (SKILLS.md) on boot.
Context Switching
Dynamically pulls project-specific memory from .liz/<project>/ without contaminating context with unrelated codebase state.
Knowledge Structure
Enforces YAML frontmatter schema (type: project_context) and progressive disclosure index.md across all concept files.
Skill Execution
Invokes sandboxed Python skills (e.g. ai_tutor, report_generator, chat_exporter) within dedicated virtual environments.
Memory Persistence
Atomically synchronizes milestones, active bugs, and remaining tasks back to the Centralized Cortex and dumps chat sessions.
CENTRALIZED CORTEX FILE STRUCTURE
~/liz/
├── .liz/ # Centralized Cognitive Cortex (OKF v0.1)
│ ├── PROJECTS.md # Master Projects Registry (Source of Truth)
│ ├── WORKSPACE_RULES.md # Operating governance & ADHD communication protocol
│ ├── SKILLS.md # Custom tool bridges & execution policies
│ ├── index.md # Root OKF bundle entrypoint (okf_version: "0.1")
│ ├── <project-name>/ # Isolated project memory bank
│ │ ├── index.md # Progressive disclosure directory index
│ │ ├── context.md # Long-term domain architecture & memory
│ │ ├── current-goal.md # Active milestone under execution
│ │ ├── progress.md # Chronological accomplishment ledger
│ │ ├── state.md # Git branch, environment & runtime state
│ │ ├── active-bugs.md # Quarantined bug register
│ │ └── remaining-features.md # Architectural roadmap
│ └── skills/ # Sandboxed agent skills and virtual environment
└── chats/ # Exported structured session transcripts