The why of your organisation, where people can review it and agents can read it.

Your tracker knows what is being built. Your code knows how. Why lives in documents, meetings and people's heads. People cope. Agents cannot.

Org Context Spec is an open specification for writing it down: plain Markdown, in a Git repository you own.

Read the specification Starter template · Worked example

A context repository

org-context/
  README.md        the agent's entry point: reading order + writing rules
  organisation/    constraints, goals, principles, ways of working
  products/        per product area: vision, effect goals, who it is for
  teams/           the bridge: team → tracker → product areas → systems
  system/          thin pointers into the engineering layer
  method/          your own rules for how content is created

A few pages, written to be read in full — not an archive to search.

Four rules

External objects are referenced, not copied. Issues stay in the tracker, goals in the goal system: ref: jira:PARK-231.
The engineering layer is a map, not content. Architecture lives in the code repositories; the context only points to it.
Thin and honest beats thick and aspirational. Write what is true today; mark what is unverified as unverified.
The model recommends, a human decides. Agents propose changes as pull requests; they never write truth directly.

Isn't this just our wiki with a connector?

A wiki connector gives an agent search over everything anyone ever wrote — current and stale, decided and merely discussed, with nothing to tell them apart. A context repository is the other end: the few pages the organisation has decided are true, every change approved by a person. They work together: the context says what holds and points to where the detail lives.

Tools are optional

Any agent that can read files can use a context repository — no server, no account. When files are not enough, orgspec-mcp is the open-source reference server, also available hosted at app.orgspec.org.