code-thermostack

Converts thermonuclear code-quality review findings into an approved local Graphite child fix stack.

2|Updated Apr 2, 2026
One-click install
npx skills add https://github.com/nseng-ai/ns --skill code-thermostack-nseng-ai
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: code-thermostack
Source: https://github.com/nseng-ai/ns/tree/main/skills/internal/code/code-thermostack
Command: npx skills add https://github.com/nseng-ai/ns --skill code-thermostack-nseng-ai

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? After a harsh maintainability review of a Graphite stack, engineers face the tedious and error-prone work of triaging findings, ranking them by risk, and turning them into clean, reviewable follow-up branches without touching the original change or mutating remotes. ## Core Features & Use Cases - Structured Review Collection: Dispatches a review-only subagent that runs the thermonuclear review definition against the stack diff and returns structured findings with evidence, risk, and validation hints. - Confidence-Ranked Batching: Ranks findings into trunk-likely, likely, uncertain, and speculative buckets, then batches them into independently reviewable branches named <base>/thermo-<NN>-<batch>. - Approval-Gated Implementation: Requires explicit user approval of a preview before any branch, commit, or stack mutation, then implements one branch at a time with validation and stop-on-ambiguity rules. - Use Case: After landing a large Graphite stack, run Thermostack to turn a brutal code-quality review into an ordered local child stack of fixes, each validated and committed, while the original branch stays untouched and nothing is pushed. ## Quick Start Ask the agent to run Thermostack on the current Graphite stack to turn the thermonuclear review findings into an approved local child fix stack.

Frequently Asked Questions about code-thermostack

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I turn code review findings into a Graphite fix stack?

Run Thermostack on the current Graphite stack. It collects structured review findings, ranks them by merge confidence, previews a batched branch plan, and after your explicit approval creates one local child branch per batch with validation and commits.

What is a thermonuclear code review in a Graphite workflow?

It is an aggressive maintainability review defined in .ns/reviews/thermonuclear-review/review.md, targeting structural regressions, branching complexity, boundary issues, and decomposition problems across the full stack diff from the stack base to HEAD.

Does Thermostack push branches or submit PRs to GitHub?

No. Thermostack is strictly local-only: it never submits PRs, pushes, lands, closes GitHub state, or mutates remotes, and it never amends the original branch. All fixes live as local child branches above the current stack tip.

Can Thermostack run without subagent support in my harness?

Yes, but only as a degraded fallback. If the harness lacks review or editing subagents, the review or implementation runs inline and the output is explicitly labeled as degraded confidence in the final report.

What happens when a fix fails validation or is ambiguous?

Thermostack stops at the first non-confident decision or validation failure. It preserves the completed clean prefix of branches, leaves state visible, and reports the exact blocker plus any pending batches rather than improvising.