Agentic Workflow & Tech Stack

If I got a new computer tomorrow and lost everything, I could clone this repo, run setup, and accurately recreate all my patterns and ways of developing software. This is a portable rebuild-my-machine scaffold — and a progressively-disclosed memory system for the AI agents that work alongside me.

What this is

A three-tier scaffolding + agent-memory system for filesystem-based AI agent workflows. Not a knowledge base itself — a system that structures how knowledge work and project development happens, and that AI agents progressively disclose into context as the work gets more specific.

01 · Kernel

Universal, forkable scaffold. Philosophy, patterns, parametric templates, deterministic scripts. Anyone can adopt this.

Explore the kernel →

02 · Stack

Opinionated toolkit — Claude Code + OpenCode + Obsidian + Zellij + Tailscale. Fork-worthy for similar setups. “Dotfiles for agentic development.”

Explore the stack →

03 · Work

Personal project patterns, agent-consumable preferences, rebuild-my-machine guide. The case study that demonstrates the kernel works.

Explore my work →

Why three tiers?

The three tiers are organizational sections, not strata themselves. They separate how portable a section is so each can be shared, forked, or kept private on its own lifecycle:

• Tier 1 — Kernel: universal claims about knowledge work. Forkable as-is.
• Tier 2 — Stack: opinionated patterns assuming my tooling. Forkable with adaptation.
• Tier 3 — Work: parametric templates of my workflow. A worked example for forks, not a payload to inherit verbatim.

What’s NOT in any tier is live state — the actual running machines, Obsidian vault, Claude sessions. The portable scaffold tops out around stratum 3 (parametric templates); stratum 4 (deterministic scripts) and stratum 5 (live instance) are generated when you run the scaffold.

Three orthogonal classification axes apply to every page. The one-sentence crib:

Tier asks who. Stratum asks how portable. Tier-of-abstraction asks what’s assumed.

Axis	Answers	Where
Tier (1 / 2 / 3)	WHO is this for?	The section it lives in (kernel / stack / work)
Stratum (1–5)	HOW MUCH of this transfers if copied?	Five strata — colored badge under each page title
Tier of abstraction (0 / 1 / 2)	WHAT substrate does this claim assume?	Tiers of abstraction — zero-tier = paper+pencil, tier-1 = digital, tier-2 = LLM-agentic

See Philosophical Alignment for the thinkers and traditions behind the claims (Thomistic / virtue-ethics lineage for moral framings; Alexander, Drucker, Simon, Wittgenstein for the structural ones).

Both classification axes are lenses, not taxonomies — real artifacts slide between strata and tiers. The labels help agents weight “how stable is this advice” without pretending certainty.

Progressive disclosure in practice

An AI agent working with a user of this scaffold progressively discloses into context:

Agent starts a task

↓ always

1Load Kernelalways

Stable, small, cache-friendly. Capture→work→output, progressive disclosure, five strata, tier-of-abstraction, SEACOW.

Tier 1 · kernelStratum 1–2Tier-of-Abstraction 0

↓ if setup matches (WSL + Claude + Obsidian)

2+ Stackconditional on toolkit

Opinionated patterns. “Use Zellij for sessions, Portagenty to launch, the WezTerm upgrade config if they want clone-tab-at-cwd.”

Tier 2 · stackStratum 2–3Tier-of-Abstraction 1–2

↓ if project type recognized (e.g. Astro+Starlight docs site)

3+ Workconditional on project + user

Parametric templates of the user’s workflow. “Apply the Astro+Starlight+Flexoki pattern, use these bashrc aliases, reference memory/tool-picks.md.”

Tier 3 · workStratum 3Tier-of-Abstraction 2

↓ agent scaffolds

✓Scaffolds in the user’s patterns

Everything up to here is inside the scaffold repo. The agent’s output at this stage is still a proposal / a plan / a new file — not running state.

inside the scaffold repo ↑↓ generated when run

▸Running the workoutside this repo

Deterministic scripts execute, live state accrues — on a specific laptop, with specific SSH sessions, inside a specific Obsidian vault, across specific Claude conversations.

outside repoStratum 4 + 5Tier-of-Abstraction 2

Start here

• Principles & Foundations — ten claims about how structured cognitive work actually works. The DNA.
• Philosophy — the condensed version.
• Kernel Reference — what the scaffold contains.
• Tech Stack — my opinionated toolkit.
• My Work — how I apply all of the above.

Status

This site is under active construction. The Principles & Foundations section is stable; tier content in Stack, Work, and rebuild guides is being populated. The repo itself is private during a PII audit pass; the live site will become publicly visible once audit is complete.