When does the tool start fixing things?

It won't fix anything until you explicitly confirm which issues to address.

What if no PR is found?

I'll tell you and suggest pushing your branch first.

How are issues prioritized?

Based on presence in code, severity, effort required, and risk of regressions; results are a per-issue Fix/Skip/Discuss recommendation.

review-and-fix-issues

Scanned

npx machina-cli add skill mrge-io/cubic-claude-plugin/review-and-fix-issues --openclaw

Files (1)

SKILL.md

2.5 KB

Review and Fix Issues

This skill fetches all AI code review issues from cubic for the current PR, investigates each one against the actual code, and reports back to the user with a prioritized assessment before fixing anything.

When to Activate

User says "check cubic comments", "cubic issues", "cubic feedback", or "cubic code review"
User mentions fixing review comments or addressing feedback
User is on a feature branch with an open PR
User asks what cubic found or what needs to be fixed

How to Use

Step 1: Communicate the plan

Tell the user:

I'll fetch all the issues cubic found on this PR, investigate each one, and report back with which ones are worth fixing. Give me a moment.

Step 2: Gather issues

Detect the current repository from git remote: git remote get-url origin
Detect the current branch: git branch --show-current
Find the PR number: gh pr view --json number --jq .number
Call get_pr_issues with repo, pullNumber, and owner to get all published issues

Step 3: Investigate each issue

For every issue returned, read the relevant code at the flagged location and assess:

Is the issue still present in the current code, or was it already addressed?
Is it a real problem (bug, security, correctness) or a stylistic nitpick?
How much effort would it take to fix?
Could fixing it introduce regressions?

Step 4: Report back

Present a summary table to the user with your recommendation for each issue:

Fix — Real problem, worth addressing
Skip — Nitpick, already addressed, or not applicable
Discuss — Ambiguous, needs user input before deciding

Group by recommendation. For each issue, include: file, line, severity, one-line summary, and your reasoning.

Step 5: Wait for the user

Do NOT start fixing anything until the user confirms which issues to address. The user decides.

Presentation

Lead with the summary table so the user gets the full picture immediately
Be concise in reasoning — one sentence per issue is enough
If all issues are already addressed, say so and congratulate the user
If no PR is found, tell the user and suggest they push their branch first

Source

git clone https://github.com/mrge-io/cubic-claude-plugin/blob/main/skills/review-and-fix-issues/SKILL.mdView on GitHub

Overview

Review and Fix Issues fetches all AI code review issues from cubic for the current PR and analyzes them against the actual code. It then reports back with a prioritized assessment of what’s worth fixing before any changes are made.

How This Skill Works

It detects the repo and current branch, determines the PR number, and retrieves all published cubic issues via get_pr_issues. For each issue, it checks if the issue remains, assesses whether it's a real problem or a nitpick, estimates effort, and considers potential regressions, then presents a concise, prioritized summary to guide next steps. It will not apply fixes until you explicitly approve which issues to address.

When to Use It

User says 'check cubic comments'
User says 'cubic issues' or 'cubic feedback'
User says 'cubic code review'
User asks to fix issues or address review
User is on a feature branch with an open PR
User asks what cubic found or what needs to be fixed
User wants a plan before applying changes

Quick Start

Step 1: Plan - I'll fetch all cubic issues on this PR and report back
Step 2: Gather - identify repo, current branch, PR number with git/gh
Step 3: Evaluate - inspect each issue and present a Fix/Skip/Discuss table; wait for your confirmation

Best Practices

Confirm repo and PR context before acting
Prioritize real bugs over stylistic nitpicks
Assess impact, effort, and risk of regressions
Summarize each issue with a one-line rationale
Wait for explicit user approval before applying fixes

Example Use Cases

PR on a feature branch with cubic comments on the login flow
Cubic feedback highlights a security nitpick in the auth logic
Multiple cubic issues on a UI refactor; determine which to fix
Cubic code review flags style issues already addressed elsewhere
No PR is detected and you're guided to push the branch

Frequently Asked Questions

Add this skill to your agents