Review
Run a resumable multi-pass review, and check a verdict.
Review is a sequence of steps with a session id rather than one long command, so a review of a large change can be picked up where it stopped.
review
Drive a multi-pass pull-request review as a sequence of resumable steps, so a long review survives a lost session.
npx mewkit review prepare <target> [--remote <name>] [--json]
npx mewkit review read <target> --session <id> --as <persona>
npx mewkit review coverage --session <id> [--json]
npx mewkit review compose --session <id> [--json]
npx mewkit review submit --session <id> [--reply] [--confirm <value>] [--json]
npx mewkit review cleanup --session <id> [--json]| Subcommand | What it does |
|---|---|
prepare | Fetch the target and open a review session |
read | Read the diff as one named persona |
coverage | Report which files and personas have been covered so far |
compose | Assemble the collected findings into a single review |
submit | Post the review; --confirm guards the irreversible step |
cleanup | Close the session and remove its working state |
Every subcommand after prepare takes the --session id that prepare returned. That id is
the whole point: reviewing a large PR across several sittings does not restart from nothing.
verdict-gate
Gate 2 proof-bundle check. Reads a machine-readable review verdict and exits 0 (PASS / PASS_WITH_RISK or a tolerated legacy markdown-only verdict), 1 (BLOCKED / invalid / missing), or 2 (usage error).
npx mewkit verdict-gate <slug | path-to-verdict.json>