Oko/Docs

Command reference

oko person assess

Has the model assess each operator turn of one session the register does not hold yet, on every dimension the product declares: whose words they are, whether an agent can act on them without guessing, whether they say when the work is done, how much of the method they leave to the agent, and whether they decide something and say why.

Invocation#

Command
oko person assess (--session ID | --transcript PATH [--session ID]) [--json]

Inputs and options#

  • --session names a session whose live transcript is on this machine; --transcript names a transcript file, whose name gives the session unless --session names it.
  • The dimensions and their choices are declared in prose in rust/src/person/messages/assessment.json; no word list decides a choice.
  • The person register is ~/Library/Application Support/Oko/person.json; OKO_PERSON_REGISTER moves it. It quotes the operator verbatim and is published nowhere.

Output and state effect#

  • Asks Brama's typed decision route once per turn, all dimensions in one request with the whole message as the state; where `oko model-service set` declared an OpenAI-compatible service, the same choices as its one `record_assessment` tool call. Each answer is kept with the route or model that read it, the least confidence it stated and the time. One turn of one session is one assessment: a turn already held with the same words is kept as it is and asks the model nothing.
  • Prints `session`, `turns`, `assessed` (each new assessment), `kept` and `unassessed` — each turn the model refused, with its answer. A refused turn does not stop the turns after it: the refusal belongs to that message.
  • The register also fills without this command: every reading pass — `oko tasks read`, and the reading stage of every `oko tasks control` pass — assesses each stored operator message of the signed-in member the register does not hold yet, and answers `assessment` with `assessed`, `unassessed` and `refusal` beside its readings. A register that cannot be read or written stops that pass's assessments with `refusal`; the readings go on.

Refusals and safe exits#

  • `oko person assess requires --session <id> or --transcript <path>`, `oko person assess does not take <flags>; it takes --session, --transcript, --json`, `oko person assess: --transcript takes a value; write --transcript <text>` and `oko person assess: the transcript name <name> carries no session id; name the session with --session` (exit 2).
  • `oko person assess: session <id> has no live transcript on this machine; name its file with --transcript`; `oko person assess: <path> holds no operator turn to assess`; `<path> cannot be read: <cause>`.
  • `oko person assess: <n> of <m> operator turn(s) of <session> could not be assessed, the rest are kept: #<turn>: <the model's answer>; …` after printing the answer, when the model refused a turn; the model's answer is `<route> refused the assessment with HTTP <status>: <body>`, `<route> answered no <dimension> choice this assessment offers (<choices>): <body>` or the unreachable gateway's own detail.
  • `turn <n> of session <id> already holds the assessment <id> of other words (<words>); name the session these words were said in`.

CLI reference#