Skip to content
Get packs

Guide · Agent skill

How to use the Chained PR Splitter agent skill (chained-pr)

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

The chained-pr skill is an agent skill for AI coding agents that splits a large code change into a sequence of smaller pull requests that depend on each other. It is for developers and maintainers who want reviewers to see focused diffs instead of one oversized PR. After this guide you will know how to install the skill, how to ask for it in plain language, and what the agent will produce.

Chained PR SplitterGentleman-Programming/gentle-ai · MIT

Quick start

  1. 1Download and unzip. Download the chained-pr ZIP from the directory page and unzip it so you have a chained-pr folder containing SKILL.md.
  2. 2Place the skill folder. Copy the folder to ~/.claude/skills/ for personal use, or to .claude/skills/ inside a project. Alternatively, upload the ZIP under Settings → Capabilities → Skills in the Claude apps.
  3. 3Make a large change. Work on a feature or ask for one that is expected to pass about 400 changed lines.
  4. 4Ask for a split. Use a plain request such as 'split this into stacked PRs' so the trigger terms match.
  5. 5Pick a strategy and review the plan. Choose Stacked PRs to main or Feature Branch Chain when the agent asks, then check the PR order, dependency diagram and review budget before it opens PRs.

What the chained-pr skill makes the agent do

When a planned pull request may exceed 400 changed lines, the skill tells the agent to slice the work into chained PRs. Each PR is one deliverable work unit, and it states its start and end, prior dependencies, follow-up work and out-of-scope items.

The agent first estimates changed lines and identifies independent work units. It then asks which chain strategy to use if none has been chosen and the budget is exceeded. Every child PR gets a dependency diagram with the current PR marked 📍, and the agent reports review budget as additions plus deletions.

  • Estimates the changed-line count before opening PRs
  • Asks for a strategy: Stacked PRs to main, or Feature Branch Chain with a draft tracker PR
  • Adds Chain Context to each PR without replacing your repo's PR template
  • Verifies each PR on its own: CI, tests, docs, rollback scope and a clean diff

Install the chained-pr skill

Download the chained-pr ZIP from the directory page and unzip it. The skill folder contains SKILL.md and references/chaining-details.md. Place the folder in one of these locations:

For all your projects, put it in ~/.claude/skills/. For a single repository, put it in that project's .claude/skills/ folder. You can also upload the ZIP in the Claude apps under Settings → Capabilities → Skills.

  • Personal: ~/.claude/skills/chained-pr
  • Project only: .claude/skills/chained-pr inside your repository
  • Claude apps: Settings → Capabilities → Skills, then upload the ZIP

Trigger the skill with natural requests

You do not need a special command. The skill loads when your request matches its trigger terms: PRs over 400 lines, stacked PRs, or review slices. Ask in plain words that describe the situation.

If the agent is already working on a large feature, you can also ask it to keep reviewer load under control. The skill also loads when an ODD feature's changed-line count goes over about 400.

  • Split this change into stacked PRs, it is around 900 lines.
  • This feature will be over 400 lines. Break it into review slices.
  • Make a feature branch chain for this work with a tracker PR.
  • Check whether my current branch needs to be split for reviewer load.

Choose between the two chaining strategies

Stacked PRs to main fits work where each slice can land independently and ship in order. Feature Branch Chain fits work where the feature must integrate before reaching main. In that case, a draft, no-merge tracker PR is created from main, and child PR #1 targets the tracker branch while later children target their immediate parent.

The skill does not mix strategies once you pick one. Its reference notes that stacked PRs have simpler retarget and rebase flow, while a feature branch chain needs a tracker and strict diff hygiene.

  • Stacked PRs: revert individual main PRs if needed; partial behavior may land
  • Feature Branch Chain: nothing lands until the chain completes; revert or hold the whole feature branch
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 └── feat/my-feature-02-shared ↑ PR #3 base: feat/my-feature-02-shared └── feat/my-feature-03-slice

Handle changes that cannot fit the limit

The skill makes one honest slicing pass. The budget changes how work is sliced, never the code itself, so the agent will not delete comments, docs or tests, or compress code to get under 400 lines.

If no cohesive split fits, the agent delivers the best split, reports the overage and why it cannot shrink, and recommends a maintainer-approved size:exception. Generated, vendor or migration diffs that cannot be split cleanly are sent to a maintainer for that exception.

Safety notes: scripts and network

According to the skill's listing, chained-pr runs no scripts. It does use the network, because the branch and PR work it describes means pushing branches and creating pull requests on GitHub.

Review what the agent pushes and which base branches it targets before you let it open PRs, especially for a Feature Branch Chain. Keep the tracker PR draft until all child PRs are reviewed and integrated.

  • Requires git and a GitHub account with permission to create pull requests
  • Runs no bundled scripts according to the listing
  • Uses the network to push branches and open PRs

What people use it for

  • Splitting a large feature into a core PR, a shared-code PR and a final slice that reviewers can read in one sitting.
  • Setting up a Feature Branch Chain with a draft tracker PR for a feature that must integrate before main.
  • Documenting a PR that must exceed the limit, with a size:exception recommendation and the reason it cannot be split further.
  • Keeping a team's stacked PRs consistent, with the same chain context and dependency diagrams in every PR.

Frequently asked questions

Is the chained-pr skill free to use?

The source material for this listing does not state pricing. The repository is listed under the MIT licence, but the skill's SKILL.md metadata says Apache-2.0, so check the licence files in the repository before you rely on either.

What do I need to use the chained-pr skill?

You need git, a GitHub account, and access to create pull requests, since the skill's branch and PR steps depend on them. The skill uses the network and runs no scripts, according to the listing.

Where do I install the chained-pr skill?

Unzip the downloaded folder into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder. You can also upload the ZIP in the Claude apps under Settings → Capabilities → Skills.

What is the 400-line limit and can I change it?

The skill asks the agent to split PRs over 400 changed lines, and it counts additions plus deletions. The 400-line and 60-minute limits are hard-coded in the skill, so if your repository has different rules they may conflict and you should review them before use.

What is the difference between stacked PRs and a feature branch chain?

Stacked PRs to main let each slice land in order, so partial behavior may reach main. A feature branch chain keeps nothing landing until the whole chain completes, and uses a draft tracker PR to integrate the feature.

What should I do if the skill says a change cannot be split?

The skill delivers its best split, reports the overage and why it cannot shrink, and recommends a size:exception. A maintainer must accept that exception, and generated, vendor or migration diffs that cannot be split cleanly are sent there.

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 chained-pr if you work on large features, your team uses stacked PRs on GitHub, or maintainers want consistent PR size rules. Pick something else for small changes under 400 lines, or for repositories whose PR size policy differs from the hard-coded limits.

Download the skill