Skip to content

Context

Every worker starts from what the repository already knows: the context that applies to its path, the handover the last session left, and the decisions already made.

The failure it answers

Agents forget. What one session learned dies with its conversation, and the next one starts over.

What a worker — a model or a person — knows at the moment it begins, and what is left behind when it stops. The context documents that apply to a path, the session and its handover, the decisions and knowledge that outlive a conversation, and the procedures a worker can load: each is a file the repository owns, read through the same commands by every tool, rather than a transcript somebody hopes to find again.

Not in this domain: who may touch which paths while the work runs, which is coordination; and whether the rules the worker loaded hold, which is governance.

The features filed under it

Each feature names this domain in its own file; the domain lists none of them. A feature added with this domain is on this page at the next build, and one removed is gone.

  • Continuity

    A session ends and the work does not: what the next worker needs is a typed record checked against git, not a transcript somebody hopes to find.

    11 of 11 claims guaranteed with a named test · 7 use cases

  • Knowledge

    What was decided, why, and what was rejected is recorded where the next session reads it, and a durable decision becomes an architecture record with typed links to what it put in force.

    7 of 9 claims guaranteed with a named test · 5 use cases

  • Context

    Context is a tree, not one file: the documents that apply to a path are resolved from the root down, the nearest one adds and never silently replaces, and a worker loads what its task needs and nothing more.

    8 of 9 claims guaranteed with a named test · 4 use cases

  • Skills

    A skill is one directory under the layer; adding it is the whole act, and the command line, the website and MCP all read it from the same file.

    3 of 3 claims guaranteed with a named test · 1 use case

The failure modes its features answer

Operational moments from the Why catalogue that name a rule, a command, a capability or a claim of a feature in this domain. Nothing here was assigned.

Every failure mode it answers (32)

What is claimed

Every claim a feature of this domain names, once, with the status it declares — guaranteed, advisory, planned or rejected — and the test that settles it. Whether a recorded run supports a claim is its own page's verdict.

32 claims, 29 guaranteed with a named test

Executable use cases

Scenarios the tool runs against itself every time this site is built.

Searches the titles and descriptions of every page this site publishes.