Command reference
oko review scan
Reads every executor transcript from its retained cursor and files structural supervisor questions and every tool call of a measured low-standing executor, as the standing reflection routine does.
Invocation#
oko review scan [--json]Inputs and options#
- The executors and their supervisors from ~/.oko/session-roles.json; their OMP transcripts under each harness's transcript root; the same source-cited standing as `oko scorecard standing`. A cursor advances only after every complete line and review entry is read with its standing. An unavailable scorecard names its cause and is retried rather than classifying the executor by words.
- Each structural entry is read from the record: todo/block, question-tool call, refusal before execution, rejected stop, and a committed diff that removes an option. For a negative measured standing, each toolCall id and its arguments is additionally queued as watched, including edits, commits and claims; nonnegative standing keeps the structural sample.
Output and state effect#
- Prints {added, executors, transcripts, unread}. The queue is ~/Library/Application Support/Oko/review.json; OKO_REVIEW_REGISTER moves it, and beside a moved defect register it is <register>.review.json. A second pass over unchanged complete transcripts adds nothing. The standing routine invokes the same scan without supervisor polling; the scan does not itself suspend the next tool call.
Refusals and safe exits#
- Unreadable roles, queue or transcript sources name their path and cause. A malformed line, missing tool identity, missing scorecard source or missing session standing is unread with its cause and leaves that transcript cursor unadvanced for retry; no unmeasured call is silently marked sampled.