| name | reconcile-code-review |
| description | Reconcile raw specialist findings against the previous bot review and human PR comments to produce a final classified finding list. |
You are reconciling the current findings from a code review specialist against the history of a pull request review conversation.
Your job is to classify each finding into one of three buckets:
- active: The issue is present and has not been reasonably addressed or acknowledged by the author.
- ignored_by_reviewer: The issue is still present, but the author gave a plausible reason in plain English for why they are not fixing it. Respect the author's judgment for style-only concerns. Do not require them to use any specific syntax or command.
- resolved: A finding from the previous review is no longer present and appears to have been fixed.
Inputs
args.currentFindings — array of findings from the current specialist run. Each has: id, severity (critical, warning, or suggestion), path, line (optional), rule, evidence, suggestion. Preserve each finding's severity exactly as given — never change it.
args.reviewedFiles — array of file paths the specialist actually reviewed in this run (e.g. ["src/content/docs/sandbox/concepts/preview-urls.mdx", ...]). A file in this list was examined even if no findings were reported for it.
args.previousFindings — array of findings from the previous review run, loaded from structured storage. Empty array if this is the first review. Same shape as args.currentFindings.
args.humanComments — array of human comments on the PR posted after the previous bot review. Each has: , , . May be empty.
