Find an object of the layer without reading everything

Situation

A client wants the rule about scope, or the decision about the parser, and has the whole layer in front of it. Reading every file to find one is what makes agents slow and expensive; asking a person is what makes them wrong.

What you run

• search <text>: durable records that contain the text, across kinds
• knowledge nodes --kind <k>: the canonical objects of one kind, with identity
• prompt list: the prompt assets, also served as resources
• over MCP: majordomus_search, majordomus_get by majordomus:// URI, majordomus_list by kind

Scenario

setup: active-task-records
given:
  - 'an active task that has already produced a checkpoint, a decision and an open question'
steps:
  - id: search
    run: ['search', 'parser']
    note: 'durable records matched literally, across kinds, without an index'
    expect:
      exit: 0
      stdout_contains: ['^decision ', 'parser', 'match']
  - id: nodes
    run: ['knowledge', 'nodes', '--kind', 'rule']
    note: 'one node per canonical object of one kind, with its identity'
    expect:
      exit: 0
      stdout_contains: ['^rule ']
  - id: prompts
    run: ['prompt', 'list']
    note: 'the repository’s own prompt assets, the ones an MCP client can also read as resources'
    expect:
      exit: 0
      stdout_contains: ['^continue ', '^debug ']
then:
  - 'the shell and the MCP server answer from the same files'
  - 'a majordomus:// URI resolves the same way through the resource read, the tool and the HTTP route'
  - 'no index has to exist for the literal search to work'

Outcome

One question, one answer, from the shell or from the server, without reading the layer end to end. The URI a client gets back resolves the same way everywhere.