A demanding review of branch structure, ownership and state models

Thermo-Nuclear Review

Find a Structure That Removes Complexity

Give an AI a focused maintainability rubric: trace new branches and unclear boundaries, assess file growth, and look for structural changes that remove whole layers of complexity.

By Cursor

Thermo-Nuclear Review: Find a Structure That Removes Complexity project image

Environment for the first run

  • Cursor Agent
  • Local Git project · explicit invocation
Repository stars
14,637
Checked
Categories
Engineering

Ask which complexity can actually disappear.

Provide a base, the changed files and the behavior that must remain. Trace findings back to ownership and the data model, then compare what each proposal removes and introduces before choosing an implementation step.

Pick a task and get started

Add this rubric to the current project, confirm that Cursor discovers it, then choose a scoped review. The default examples return a report so you can inspect the evidence before selecting a change.

What decision do you need from this review?

Check whether a working branch makes the surrounding code harder to maintain.

  1. Add this one skill to your project

    For the first run, save the pinned SKILL.md at .cursor/skills/thermo-nuclear-code-quality-review/SKILL.md in your project. Keep its YAML frontmatter and Team Kit's LICENSE, then select the skill explicitly in Cursor's / menu. Bundle options are explained below.

    Read the author’s setup guide

    First run: confirm discovery, then review committed changes

    Open the pinned SKILL.md linked on this page, use Raw to save the complete file, and place it at .cursor/skills/thermo-nuclear-code-quality-review/SKILL.md under your project root. Keep the YAML frontmatter and save Team Kit's LICENSE alongside it. This skill directory currently contains only SKILL.md, with no extra scripts or references. Reopen the project and confirm it appears in Customize → Skills. In Agent chat, type / and select thermo-nuclear-code-quality-review, then send the task below. The selection attaches to that message; the file sets disable-model-invocation: true, so invoke it explicitly. If it is missing, check the folder nesting and filename. Start from a branch with real changes and an available main reference. Check the returned locations, before/after evidence and coverage gaps.

    Read Cursor's directory and invocation guide
  2. Give your AI this first task

    These editorial review tasks apply the author's rubric. Replace bracketed inputs. The slash line belongs in Cursor chat, not a terminal. Report fields, scope limits and read-only boundaries are choices in these examples; we have not run them on a project.

    /thermo-nuclear-code-quality-review Review the committed changes on the current Git branch against main. First confirm that main and HEAD exist, record the IDs of main, HEAD and their merge base, and report the working-tree state. Use main...HEAD: the diff from that merge base to HEAD. List staged, unstaged and untracked files separately and exclude them from this pass. If main is unavailable or no merge base can be found, explain the missing context and ask me for an available base; if there is no diff, say so. Read the full changed files, relevant callers, existing types and project conventions at HEAD; use the merge-base versions for the before side. Keep uncommitted working-tree content out of the evidence. Identify any unread or out-of-scope material. Separate problems introduced by this change from pre-existing issues, and trace effects across module boundaries. Return a small set of high-confidence findings in the skill's order: structural regressions, major simplification opportunities, branching growth, boundary and type-contract problems, decomposition, modularity, then readability. For each finding, give the file and location, code evidence, maintenance impact, an actionable remedy and the behavior it must preserve. Cite the definition and uses of any existing helper you recommend. Keep unverified hypotheses separate. Return a review only; do not edit files, commit or push. If there is no clear issue, say that none was found within the inspected scope. List verification that was not run, and do not present the review as a functional or security test pass.

  3. Check what you receive

    • A review with the base, commit IDs, inspected material and coverage gaps.
    • A short, evidence-backed finding list ordered by structural impact, with simplification directions and behavior to preserve.
    Read the pinned review rubric
More about this skill

Thermo-Nuclear Code Quality Review comes from Cursor's official plugins repository. It asks how a change affects the next person who must maintain it: whether responsibilities have become entangled, feature exceptions have entered shared flows, state is tested repeatedly, or helpers merely forward arguments. Its central idea is code judo: change the model or ownership so that a family of conditions, modes or wrappers can disappear.

The rubric puts structural regressions and major simplification opportunities ahead of branching, boundaries, types, decomposition and readability. Taking a file from below 1,000 lines to above 1,000 demands an explanation. The author allows an exception when there is a compelling structural reason and the file remains well organized. Objections to unknown, sequential calls or optional fields also depend on the actual context.

The intended result is a small set of high-confidence, actionable comments. A companion subagent can receive a diff and full file contents collected by a parent agent. The standalone skill itself is a Markdown rubric, with no scanner, test runner or merge gate. Establish the scope, inspect the evidence, then decide which refactoring proposal to implement.

Sources and verification

Sources checked

Prepared from Team Kit's rubric, subagent and plugin material at commit 4dd1d058, plus Cursor's current skills documentation and Git's diff reference. Source bytes and the two rubric copies were checked. We did not install a plugin, run a review or test a refactor.