Guide · Agent skill

How to use the Chained PR Splitter agent skill

Updated 2026-11-02 · written with AI from the project's own documentation and checked against our catalogue data

The Chained PR Splitter is an agent skill from the gentle-ai project (skills/chained-pr, MIT licence) that teaches an AI agent to break a large code change into a sequence of smaller, dependent pull requests. It is for developers and teams whose AI-assisted changes keep producing pull requests that are hard to review. After this guide you will know what the skill changes about the agent's behaviour, how to install it, how to trigger it with plain requests, and what it needs before you rely on it.

Chained PR SplitterGentleman-Programming/gentle-ai · MIT Download the skill

Quick start

1Download and unzip. Download the skill ZIP from this page and unzip it into ~/.claude/skills/ or your project's .claude/skills/ folder.
2Or upload in the Claude apps. In the Claude apps, open Settings → Capabilities → Skills and upload the skill.
3Confirm the files. Check that SKILL.md and references/chaining-details.md are present and that the skill name is gentle-ai-chained-pr.
4Ask for a split. Describe the change and mention that it is over 400 lines or that you want stacked PRs or review slices.
5Pick a strategy. When asked, choose Stacked PRs to main for independently landable slices, or Feature Branch Chain when the feature must integrate first.
6Review the output. Check the strategy, PR order, current PR boundary, dependency diagram, review budget and verification plan the agent returns.

What is an agent skill, and what does this one make the agent do?

An agent skill is a folder of instructions and reference notes that an AI agent loads when a task matches the skill's description. This skill is a set of instructions; it does not change code by itself. The agent does the work and follows the skill's rules while doing it.

When a planned pull request may exceed 400 changed lines, or when you ask for chained or stacked PRs, review slices, or reviewer-load control, the skill tells the agent to split the work. Each resulting PR should be reviewable in about 60 minutes or less and should contain one deliverable work unit with its tests and docs.

Estimate changed lines and identify independent work units before opening any PR.
Choose a strategy: Stacked PRs to main, or a Feature Branch Chain with a draft tracker PR.
State start, end, prior dependencies, follow-up work and out-of-scope items in each chained PR.
Include a dependency diagram in every child PR, marking the current PR with a pin symbol.
Report the review budget as additions plus deletions, along with the verification plan.

Install the skill with a ZIP download or the Claude apps

Download the ZIP for this skill from this page. For a personal install, unzip it into ~/.claude/skills/. For a single project, unzip it into the project's .claude/skills/ folder so it applies only to that repository.

If you use the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills.

Personal install: unzip into ~/.claude/skills/
Project install: unzip into .claude/skills/ inside your repository
Claude apps: Settings → Capabilities → Skills, then upload

Check the files after installing

The skill folder should contain SKILL.md, which holds the activation rules, hard rules, decision gates and output contract, and references/chaining-details.md, which holds strategy diagrams, the PR body section, branch commands and reviewer guidance. The agent reads the reference file when it needs those details.

Open SKILL.md and confirm the name is gentle-ai-chained-pr before relying on it.

SKILL.md: main instructions
references/chaining-details.md: strategy notes and branch commands

How to trigger the skill with natural requests

You do not need a special command. Describe the situation in plain language and mention size, stacking or review load. The skill's activation contract covers requests about PRs that may exceed 400 changed lines, and requests for chained or stacked PRs, review slices or reviewer-load control.

If a planning tool has already flagged a 400-line budget risk or recommended chained PRs, the skill also loads on that signal. If you have not yet chosen a strategy and the budget is exceeded, the agent is instructed to ask you which one to use.

"This change will be over 400 lines. Split it into reviewable PRs."
"Make this a stacked PR series to main, one slice per PR."
"Set up a feature branch chain with a tracker PR for this feature."
"Keep reviewer load low: split this refactor into review slices."

Choose between Stacked PRs and a Feature Branch Chain

Stacked PRs to main suit work where each slice can land independently and in order. Rollback is per PR, but partial behaviour may reach main. A Feature Branch Chain suits features that must integrate before reaching main. Nothing lands until the chain completes, and the agent must create a draft, no-merge tracker PR. Child PR #1 targets the tracker branch and later children target their immediate parent.

The skill also says not to mix strategies once you have chosen one, and to treat polluted diffs as base bugs to fix by retargeting or rebasing.

main
 └── feat/my-feature              ← tracker/final integration branch
      ↑ PR #1 base: feat/my-feature
      └── feat/my-feature-01-core
           ↑ PR #2 base: feat/my-feature-01-core

Know what the skill needs and what it does not do

The skill needs Git and a hosted pull request workflow. Its branch and PR commands assume GitHub-style pull requests. The agent also needs repository access and permission to create branches and pull requests if you want it to do that work.

The skill runs no scripts and uses no network access, according to its listing. Because it is only instructions, the risk lies in what the agent does with your repository, so review the branches and PRs it creates the same way you would review a colleague's work.

If a change is a large generated, vendor or migration diff that cannot be split cleanly, the skill tells the agent to ask a maintainer for a size:exception rather than forcing a split.

Requires Git and a GitHub-style PR workflow
Requires repository permissions if the agent creates branches or PRs
Runs scripts: no
Uses network: no

What people use it for

Splitting a large feature into a series of stacked PRs, each landing in order on main.
Keeping a multi-part feature on a tracker branch until all child PRs are reviewed and integrated.
Breaking a refactor that touches many files into review slices with tests kept alongside the code they verify.
Documenting exception requests with a size:exception rationale when a generated or migration diff cannot be split.

Frequently asked questions

How do I install the chained PR splitter agent skill?

Download the ZIP from this page and unzip it into ~/.claude/skills/ for a personal install, or into a project's .claude/skills/ folder. In the Claude apps you can instead upload it under Settings → Capabilities → Skills.

Is the gentle-ai-chained-pr skill free to use?

The skill is listed under the MIT licence on this page, which is the licence stated for the source repository. Check the licence file in the repository for the full terms before redistributing it.

What does the skill require?

It requires Git and a hosted pull request workflow, since its branch and PR commands assume GitHub-style PRs. The agent also needs repository access and permission to create branches and pull requests if it is to do that itself.

Does the skill run scripts or use the network?

According to its listing, it runs no scripts and uses no network. It consists of instructions and reference notes that guide the agent.

Can I use it if my team does not use pull requests?

No. The skill is designed for branch-based review on a hosted PR workflow, and it is not suitable for repositories that do not use GitHub-style pull requests or branch review.

How is this different from just asking the agent to split a PR?

The skill gives the agent fixed rules: a 400-line limit, a 60-minute review target, a choice of two defined strategies, required dependency diagrams and a draft tracker PR for feature chains. Without it, the agent has no consistent structure to follow.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Use this skill if AI-assisted changes regularly produce pull requests too large to review and your team works with stacked or chained PRs on a hosted Git workflow. Skip it for small, focused changes under 400 lines, for repositories without branch-based pull request review, or for large generated, vendor or migration diffs that cannot be split cleanly.

Download the skill