Iwo Szapar · personal operating system

    THE SECOND BRAIN
    ARCHITECTURE

    Context first. Durable records second. Agents only where authority, checks, and ownership keep them truthful. This is the working system behind the work - not a chat history, not a dashboard, and not one repository.

    GIT + SUPABASE ARE SOURCES OF RECORDGITHUB ISSUES OWN LIVE EXECUTIONDERIVED INDEXES ARE NEVER PROOF
    HUMANS

    IWO + COLLABORATORS

    Direction, authority, decisions, review, customers, and the real-world work.

    INTERFACES

    CODEX · CLAUDE CODE · MCP

    Execution surfaces read the workspace contract and load scoped context before they act.

    ROUTING

    RULES → SKILL → SOURCES

    The request selects an owner workflow and only the sources needed to answer it.

    01

    FOUNDATION

    FOUR CONNECTED LAYERS

    LAYER 1

    VERSIONED CONTEXT

    Git-backed instructions, product truth, runbooks, source records, code, and change history.

    LAYER 2

    DATA + MEMORY

    Supabase holds MemoryOS and operating records: structured, searchable, dated, linked, and owner-scoped.

    LAYER 3

    SKILL ROUTING

    142 owner skills and rules turn requests into bounded, repeatable operating workflows.

    LAYER 4

    AUTOMATION + EGRESS

    Vercel, launchd, website, Factory, email, LinkedIn, finance, and approval-gated output rails.

    02

    REPOSITORY BOUNDARIES

    THE SYSTEM IS CONNECTED, NOT MONOLITHIC

    REPO 01

    IWOSZAPAR.COM

    Marketing site, questionnaire, local purchase records, admin systems, fulfillment bridge, and operating data.

    REPO 02

    HEALTH CHECK MCP

    The local npm package that checks and maintains MemoryOS infrastructure.

    REPO 03

    FACTORY REMOTE MCP

    Creator accounts, v2 Stripe Connect checkout, remote tooling, and repo-generation delivery.

    REPO 04

    FACTORY GTM CONTEXT

    Separate creator-GTM and Factory-ops repository. It is related, but deliberately outside this repo.

    03A

    AUTHORITATIVE CONTEXT

    FILES THAT EXPLAIN THE WORK

    Files hold the exact instructions and the detailed evidence. They are versioned, reviewable, and readable without an agent.

    AGENTS.md + CLAUDE.mdThe operating contract, product boundaries, safety rules, and deep workspace context.
    .claude/rules + docsProgressive disclosure: domain rules, runbooks, architecture, migrations, and gotchas.
    memory/semantic/Versioned structural and product documentation - not a dumping ground for lived knowledge.
    GitHub IssuesCanonical active work: backlog, blockers, decisions needing input, and PR-linked execution.
    Git history + PRsThe reviewable change record and the exact state of versioned work.
    Rule: do not replace evidence with an unsupported summary. Follow the source path.
    03B

    OWNED DATA PLANE

    MEMORY + OPERATING RECORDS

    Supabase is both the durable memory layer and the operating database. Records have a retrieval path, identity, ownership, history, and controls.

    memory_knowledgefacts, patterns, insights, lessons, references, confidence, decay, and full-text/vector search
    memory_decisionscontext, options considered, rationale, chosen path, and outcomes
    memory_sessionssession summaries, skills used, files changed, and branch history
    knowledge_linkstyped support, contradiction, derivation, supersession, and prerequisite links
    operational dataprospects, purchases, content, newsletters, finance, tasks, CRM activity, and system health
    Rule: preserve corrections, provenance, uncertainty, tenant boundaries, and historical state.
    04

    REUSABLE EXECUTION

    OWNER SKILLS TURN CONTEXT INTO WORK

    SKILL 01

    /overview + /next-task

    orient the day and surface the highest-leverage AI-ready work

    SKILL 02

    /begin + /end

    tie a work session to GitHub execution, a branch, verification, and learned context

    SKILL 03

    /content + channel writers

    route research through writing, evaluation, voice gates, and publishing

    SKILL 04

    /enrich-prospects + /followups

    keep relationship context, qualification, and next actions current

    SKILL 05

    /recall + /save + /grow

    retrieve, persist, validate, and improve the durable memory layer

    SKILL 06

    /accountant + /validate-build

    close finance evidence and protect delivery with explicit verification

    05

    AUTOMATION + CONTROL

    SCHEDULED SYSTEMS WITH GUARDRAILS

    RAIL 01

    VERCEL CRONS

    Email eligibility and queues, enrichment, health checks, accounting, finance, fulfillment, publishing, purchase reconciliation, and LLM budget alerts.

    RAIL 02

    LOCAL LAUNCHD

    Metrics, LinkedIn inbox sync, growth, follow-ups, approved sender rail, stats, comment performance, and SEO monitoring.

    RAIL 03

    HUMAN GATES

    Voice validation, exact content approval, reviewed replies, explicit send authority, build checks, and migration dry-runs.

    RAIL 04

    OBSERVABILITY

    cron_runs, error_log, health checks, Sentry, PostHog, finance audit invariants, and source-backed admin dashboards.

    06

    TWO LOOPS, ONE SYSTEM

    CAPTURE CAREFULLY. RETRIEVE TRACEABLY.

    CAPTURE

    01Is it durable and useful?
    02Search the authoritative source
    03Update the smallest valid record
    04Keep corrections + history
    05Verify with one retrieval

    RETRIEVE

    01Read the relevant operating rules
    02Search authoritative files + memory
    03Read the canonical source before answering
    04Prefer newer facts and linked entities
    05State source path and uncertainty
    07

    THE WORK CYCLE

    FROM SIGNAL TO A BETTER SYSTEM

    1Route
    2Retrieve
    3Plan
    4Execute
    5Verify
    6Record
    7Monitor
    COMMAND

    Priorities, decisions, GitHub-owned work, branches, and verifiable delivery.

    MEMORY

    Structured knowledge, decisions, sessions, links, freshness, and retrieval quality.

    OPERATIONS

    CRM, content, purchases, fulfillment, accounting, and approval-gated customer output.

    OBSERVABILITY

    Health, cron status, errors, usage budgets, evidence, and the next system improvement.

    THE PRINCIPLE

    The brain should make the next action easier without blurring ownership, authority, or the path back to truth.