Method / Context OS

Build the smallest system that lets learning travel.

Start with real work. Preserve evidence. Let qualified people decide. Publish context other people and agents can safely depend on.

Principles

Use your mind to think and the system to remember.
  • People decide what matters. Agents may retrieve and compare evidence; qualified owners decide what becomes shared context.
  • Preserve why. Keep the sources, difference, reasoning, and lineage behind every important release.
  • Start from outcomes. Choose two or three real workflows and work backward to their context dependencies.
  • Capture first, classify later. Let repeated use reveal the distinctions worth maintaining.
  • One shared system does not mean one central brain. Centralize standards and lineage while distributing ownership and judgment.

Five roles inside one corpus

context-os/
├── README.md   # human map
├── AGENTS.md   # agent rules
├── context/    # current approved beliefs
├── sources/    # raw evidence
├── skills/     # repeatable methods
├── work/       # drafts and proposed changes
└── releases/   # approved outputs safe to reuse

Sources preserve what happened. Context states what is currently believed. Skills describe how to perform repeatable work. Work holds change in progress. Releases identify what others may safely reuse.

Packages: the unit of shared context

Create a package only when knowledge has a distinct owner or cadence, or when another person, team, agent, or workflow needs to depend on it. Record its owner, reviewers, current release, evidence sources, consumers, permissions, cadence, and dependencies.

Separate logs from canon. Logs preserve facts such as calls, tickets, experiments, and results. Canon states the current reviewed belief or standard. Facts accumulate; an owner interprets them; canon receives a new release.

The context lifecycle

  1. Capture evidence with its source and relevant context.
  2. Decide what it means through the named owner.
  3. Propose a version with the change and reasoning visible.
  4. Review the difference before downstream work depends on it.
  5. Release an explicit approved version while preserving history.
  6. Use that release through role-appropriate access surfaces.
  7. Learn from the result and return it as evidence.

Dependencies and federation

The folder identifies where context lives. The owner identifies who reviews changes. The release identifies which version consumers can trust. Dependencies reveal which downstream work may now be stale when an upstream package changes.

That is federation: one shared system and rulebook, with judgment distributed across the people closest to the work.

Proof gates

Personal

A real workflow begins from reusable context, preserves provenance, and can be updated without rebuilding the system.

Team

A contributor proposes a real change, the owner reviews it, the team cuts a release, and another workflow begins from it.

Organization

A second team consumes the first team’s approved release without reconstructing its context. A later change exposes downstream impact. Internal owners can run the next lifecycle without the original builder.

Common questions

What is a Context OS?

A Context OS is a file-based working system that separates source evidence, current beliefs, repeatable skills, drafts, and approved releases. At team scale it adds owners, permissions, reviews, versions, dependencies, and access surfaces so people and agents can safely reuse shared context.

What is a context package?

A context package gives important shared knowledge one canonical home, named owner, review path, release history, known consumers, and explicit dependencies. Create one when context has a distinct owner or cadence, or when another team or workflow needs to depend on it.

Does everyone need to use Git?

No. Git can preserve lineage, differences, and releases underneath the system while most people use files, web, search, API, MCP, or role-appropriate interfaces. The governance model is mandatory; Git expertise is not.

How should a company start?

Choose two or three valuable outcomes, trace the shared context each requires, identify owners and consumers, and build the smallest package system that supports one complete evidence-to-release loop. Add a second team only after the first loop works.