code-review

{
  "name": "code-review",
  "version": "1.3.1",
  "description": "A diff review that reports what it could not check. Finds with fourteen named angles that are forbidden from suppressing each other, so two reasons for flagging one line both survive to be judged; verifies with three verdicts, where realistic-but-unproven stays in as PLAUSIBLE rather than being dropped with the noise, and only REFUTED leaves; then closes with a coverage ledger naming every angle, checklist and gate that never ran, and why. Learns the target repository at runtime rather than carrying a hard-coded project map \u2014 gate commands from the package scripts and CI config, frameworks from the installed dependency versions, global controls and cross-package boundaries by grep \u2014 so a candidate naming a framework the repo does not install is refuted from evidence rather than from a constant somebody wrote into a skill. Three depths, each printing its own budget line at the top of the report so what ran is visible rather than inferred; areas and ten focus lenses that compose; checklists for TypeScript, Next.js, NestJS, React Native, frontend/web, security and logic bugs, routed by path and lens; sharding above a size threshold with the fan-out reconciled against the bucket list, because a harness that loses an agent reports the wave complete. One verifier vote rather than a judge panel, on the measurement that panels saturate at about two effective votes. Read-only on source: it reports findings and does not apply fixes. Carries a token-light prepush gate on the outgoing diff.",
  "author": {
    "name": "Fledgeling",
    "email": "[email protected]"
  }
}