Public case study of a terminal-first, human-in-the-loop AI operating model.
This repo explains how I use AI agents, local memory, MCP tools, browser automation, reusable skills, and verification loops to build real software and operational workflows. It is not presented as traditional solo hand-coding. It is presented as an AI-native operating practice: I act as architect, operator, reviewer, and quality controller while agents perform bounded implementation, research, testing, and documentation work.
Many people use AI as a chat window. My work is built around a deeper pattern:
- agents with explicit roles and boundaries
- terminal-first workflows that preserve state and evidence
- reusable skills for repeatable operations
- browser automation for real-world interaction and inspection
- local memory and repo notes for continuity
- verification loops that make weak assumptions visible
- resource controls so parallel agents do not collide over shared tools
The goal is not to hide that AI is involved. The goal is to show the operating model around it.
This is a documentation-first repo. It contains:
README.md— the top-level framing for the single-person AI operating model.docs/operating-model.md— the role split between human operator, main agent, subagents, tools, and memory/docs.docs/skills-and-memory.md— how reusable skills and durable memory keep repeated AI work from restarting at zero.docs/browser-resource-governance.md— how shared browser automation resources are controlled so agents do not collide.docs/agent-safety-model.md— the credential, filesystem, approval, and review boundaries around agent work.docs/verification-loop.md— the evidence and adversarial-review loop used before claims are treated as finished.docs/public-portfolio-map.md— how this operating model maps onto the public portfolio repos.examples/session-slice.md— a sanitized example of a session structure and decision trail.
It does not contain private memory, private lead data, client data, secrets, raw logs, or the full local workspace.
| Artifact | What it shows |
|---|---|
docs/operating-model.md |
How the human/operator role, agents, tools, and review loops fit together |
docs/skills-and-memory.md |
Reusable skills, durable memory, and continuity model |
docs/browser-resource-governance.md |
Shared browser/session controls for parallel agent work |
docs/agent-safety-model.md |
Agent access boundaries, credential handling, and human approval gates |
docs/verification-loop.md |
How claims are checked before work is treated as complete |
examples/session-slice.md |
Compact example of the operating style |
I build AI-assisted systems that turn noisy real-world signals into bounded operational decisions.
That shows up in several public projects:
trading-bots: state-aware research, telemetry, risk controls, and proof surfaceswebsite-audit-engine: evidence-tiered website audit automationleadops: retrieval, mailbox parsing, lead normalization, and operator review queuessemantic-demo: interactive semantic vector visualizationopencode-fork: OpenCodex, an OpenCode fork with task DAGs, semantic retrieval, TUI proof artifacts, and stateful agent runtime
This repo is the meta-layer: how those projects get built and operated.
- Operating Model
- Skills and Memory
- Browser Resource Governance
- Agent Safety Model
- Verification Loop
- Public Portfolio Map
For AI and AI-adjacent roles, this repo is meant to show:
- AI workflow design
- prompt-to-production judgment
- agent coordination
- tool and context management
- browser automation literacy
- evidence discipline
- documentation hygiene
- product sense around operational systems
- Not a claim that I hand-wrote every line of every project.
- Not a claim that agents remove the need for engineering judgment.
- Not a dump of private workspace data.
- Not a replacement for reading the project repos themselves.
The claim is narrower and stronger: I can operate AI systems in a structured enough way to produce useful software, audits, workflows, research surfaces, and public-facing artifacts while keeping verification and human judgment in the loop.
The precise framing for this work:
- AI-native systems operator: someone who works through AI-enabled tools as the normal interface to software and operations.
- Human-in-the-loop AI systems builder: someone who designs the workflow, constraints, review gates, and outputs around AI work.
- AI automation engineer: someone who builds and operates automations where LLMs, scripts, browsers, data, and humans each have defined roles.