Why Context Is the Whole Game

A generic AI assistant gives generic advice. Personal OS is designed to give advice calibrated to your situation — your projects, your constraints, your past decisions, your preferences. That calibration comes from context you add to the system. The more relevant context your agent has, the less you have to re-explain, the fewer wrong-direction suggestions you receive, and the more useful your sessions become.

Context lives in three places: your standing instructions (CLAUDE.md), your identity file, and your knowledge base.

CLAUDE.md — Standing Instructions

CLAUDE.md is the document your agent reads at the start of every session. It contains your name, your roles, your current projects, your tools, your preferences, and any standing rules you want enforced.

The most important things to keep current in CLAUDE.md:

Active projects. If a project isn’t in CLAUDE.md (or referenced from it), your agent doesn’t know it exists. Keep your active project list current — this is the single highest-leverage update you can make.

Role context. If you wear multiple hats — founder and individual contributor, or full-time employee and side project builder — describe how they relate and what the intended split is. Your agent will use this to flag imbalances and protect your attention.

Recurring preferences. Any standing rule worth repeating should be written down once. “Always ask before scheduling anything on Friday afternoons.” “Default to async over synchronous follow-up.” “My partner’s name is spelled this way.” These accumulate into a high-fidelity profile over time.

Communication style. How formal or casual should your agent be? Should it push back aggressively or gently? Should it offer options or just decide? These defaults matter more than they seem.

To edit: open /Users/yourname/Desktop/ops/CLAUDE.md and update directly. The agent reads it fresh each session.

identity.md — Who You’re Becoming

05-habits/identity.md is a different kind of document. Where CLAUDE.md is operational, identity.md is aspirational — it describes who you’re building yourself into, not just what you’re doing right now.

Useful content for identity.md:

• Identity statements (“I am someone who…”) that anchor your habits
• Your principles and values as you’d articulate them
• The version of yourself you’re working toward, described concretely
• What “operating at your best” looks like on a typical day

Your agent references this file when you’re stuck, when you’re tempted to overcommit, or when a decision touches your deeper priorities. It’s not a vision board — it’s a compass your system actually uses.

03-resources/knowledge/ — Your Knowledge Base

The 03-resources/knowledge/ folder is where context about specific domains accumulates over time:

• Decisions: knowledge/decisions/ — past decisions with their rationale. When you make a significant call, log it here. Future sessions can retrieve it.
• Lessons: knowledge/lessons/ — what you’ve learned from mistakes, experiments, and hard projects.
• Mental models: 03-resources/references/mental-models.md — frameworks you reach for when thinking through problems.
• Playbooks: 03-resources/playbooks/ — how you approach recurring decisions and workflows.

The knowledge base grows through use. After a significant decision, your agent will offer to document the rationale. After a project completes, it will offer to capture the lessons. Say yes — this is the compounding layer of the system.

A Practical Starting Point

If you’re configuring manually rather than running /onboard, start with three things:

1. Fill in your name, roles, and active projects in CLAUDE.md
2. Write 3–5 identity statements in identity.md
3. Add one past decision to knowledge/decisions/ with its rationale

That’s enough to make your first session meaningfully better than a blank slate. Everything else can grow from there.