score-findings

What problem does it solve?

Pull requests and automated review tools generate many findings, including duplicates, vague suggestions, and false positives, which waste reviewer time and slow merges. This Skill automates verification against the actual repository code, filters out low-value findings, and prioritizes actionable fixes so reviewers can focus on real regressions.

Core Features & Use Cases

• Category-aware deduplication: Merge findings by file:line:category and surface possible shared root causes when multiple related findings appear in the same function.
• Parallel, batched verification: Spawn parallel Sonnet agents to read code context, compare branch vs origin/master, and evaluate logic or semantic regressions with up to five findings per agent.
• Deterministic scoring and filtering: Assign 0-100 scores based on accuracy and actionability and return only findings >=50 with a breakdown of filtered items for prompt tuning and triage.
• Use case: Run as part of pre-merge checks to automatically verify and prioritize code-review findings from multiple analysis agents.

Quick Start

Score the provided list of findings for the current branch and return the high-confidence findings along with counts of filtered results.