Cross-Issue Regression Sweep

SkillFiles & storage

Lets your agent scan open issues to find ones a PR could fix or conflict with, with file and line evidence.

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the Cross-Issue Regression Sweep skill

About this capability

Scan open issues to find issues that a PR can also fix or conflict with. Report each relationship with file and line evidence. Use this skill during PR review to find related fixes and risks.

What this skill tells your AI

The instructions your AI receives, as published by nvidia/nemoclaw in .agents/skills/nemoclaw-maintainer-cross-issue-sweep/SKILL.md and read by ahel’s review.

Find open issues that a PR can affect in addition to its linked issue. Report two relationship types:

  • Adjacent fix — The PR can also resolve another issue.
  • Conflict — The PR can prevent the behavior that another issue requests.

Prerequisites

  • gh CLI authenticated
  • A target repository with open issues
  • An open PR to scan

Repo policy

The defaults use NemoClaw conventions. Edit repo-policy.md for another repository.

Workflow

Copy this checklist into your response and check off each step:

Cross-issue sweep progress:
- [ ] Step 1: Extract fingerprint (files, symbols, error strings, primary issue)
- [ ] Step 2: Search candidate issues (capped at 30, primary excluded)
- [ ] Step 3: Classify each candidate (4-class with evidence)
- [ ] Step 4: Apply reverse-link boost
- [ ] Step 5: Filter (drop UNRELATED, SAME_ISSUE_DIFF, low-confidence)
- [ ] Step 6: Render report using templates/report.md

Step 1: Extract fingerprint

scripts/extract-fingerprint.sh <pr-number>

The script collects changed files, changed symbols, error strings, and the linked issue. See checks/fingerprint-extraction.md.

Step 2: Search candidate issues

scripts/search-candidate-issues.sh <fingerprint-json>

Search these three inputs. Keep no more than 30 candidates:

  • Per symbol: top 10 by recency
  • Per file path: top 5 by recency
  • Per error string: top 5 by recency

Remove duplicates and the linked issue.

Step 3: Classify each candidate

Classify each candidate with the rules in checks/relationship-judgment.md:

  • ADJACENT_FIX — The PR can resolve this issue.
  • CONTRADICTING — The PR conflicts with the requested behavior.
  • SAME_ISSUE_DIFF — same root bug as PR's primary issue (dedup filter)
  • UNRELATED — no meaningful relationship

For ADJACENT_FIX or CONTRADICTING, cite:

  • A PR diff line.
  • An issue symptom.
  • Confidence: high / medium / low

Classify the issue as UNRELATED if this evidence is not available.

Step 4: Reverse-link boost

Increase confidence by one level if the issue body or comments mention the PR number.

Step 5: Filter

  • Remove UNRELATED and SAME_ISSUE_DIFF results.
  • Remove low-confidence results.
  • Keep high- and medium-confidence ADJACENT_FIX and CONTRADICTING results.

Step 6: Render report

scripts/render-report.py < classifications.json

See templates/report.md for the format.

Reference files

Scripts (execute, do not read)

  • scripts/extract-fingerprint.sh — symbols, paths, and error strings
  • scripts/search-candidate-issues.sh — GitHub Search wrapper, dedupe, cap
  • scripts/render-report.py — report renderer

Composition with other skills

This skill is an optional follow-up to nemoclaw-maintainer-pr-comparator. The comparator does not run this skill or use its findings in the score. Run this skill when a maintainer asks for related-issue evidence. Report the evidence separately.

Limits

The skill does not:

  • run PR code against adversarial inputs
  • trace data flow with a static analyzer such as CodeQL or Semgrep
  • disambiguate symbols across codebases with a machine-learning model

Signals

GitHub stars
22k
Forks
3k
Last commit
Sep 2026
Advanced
Catalog kind
skill
Gateway key
nemoclaw-maintainer-cross-issue-sweep
Source
github.com/nvidia/nemoclaw