Agent Skills: GH PR

Writes clear pull request descriptions from the current branch diff against a base branch. Defaults to `main`, but supports `dev` or any other user-specified comparison branch. Use when the user asks for a PR description, pull request summary, or a markdown write-up for changes against a base branch.

UncategorizedID: Unique-Divine/jiyuu/gh-pr

Install this agent skill to your local

pnpm dlx add-skill https://github.com/Unique-Divine/jiyuu/tree/HEAD/ai-skills/gh-pr

Skill Files

Browse the full folder contents for gh-pr.

Download Skill

Loading file tree…

ai-skills/gh-pr/SKILL.md

Skill Metadata

Name
gh-pr
Description
Writes clear pull request descriptions from the current branch diff against a base branch. Defaults to `main`, but supports `dev` or any other user-specified comparison branch. Use when the user asks for a PR description, pull request summary, or a markdown write-up for changes against a base branch.

GH PR

Purpose

Create a clear, logical pull request description for the current branch diff against a base branch. A good PR body explains the problem, the decisions that shape the implementation, and the resulting behavior—not merely the changed files.

Instructions

When using this skill:

  1. Determine the comparison branch first:
    • Use the branch the user specifies when they name one.
    • Otherwise, default to main.
    • If the correct base branch is unclear, ask the user before drafting the pull request description.
    • If the repo clearly uses another default integration branch, such as dev, use that instead of assuming main.
  2. Read any linked issue, epic, journal, design, or acceptance criteria before drafting. Treat agreed decisions and accounting/event semantics in those documents as source material for the body.
  3. Write a pull request description that explains the rationale behind the changes, not just the file-by-file edits. State what was broken or incomplete, why the chosen boundary or data contract is correct, and which existing semantics intentionally remain unchanged.
  4. Put reviewer context in this order: the PR title and synopsis, ## Rationale, ## Key Changes, then any additional behavior-oriented sections. Keep Rationale and Key Changes unnumbered. Number only the additional topic headings as ## 1 - ..., ## 2 - ..., and so forth. The numbering improves scanning across distinct topics; it does not imply implementation order or runtime sequence.
  5. For changes with non-obvious behavior, cross-system effects, accounting, migrations, or several related fixes, use the additional sections to explain the relevant logic. Name them for the subject—for example ## 1 - Partial-close behavior, ## 2 - Borrowing settlement, or ## 3 - Historical compatibility—rather than using implementation-oriented headings.
  6. For migrations, accounting changes, data-contract cutovers, or asynchronous authority/fallback behavior, lead the relevant additional section with a compact ASCII flow or field map. Use prose afterward only for the rationale the flow cannot convey. Do not use Mermaid.
  7. Do not add a dedicated Validation, Testing, or similarly named section, and do not enumerate routine test, lint, build, formatting, or check runs. Mention test strategy or coverage only when it materially changes or creates a reviewer-relevant risk; place that context in the most relevant behavior-oriented section.
  8. Keep the writing concise and non-repetitive, but do not omit the reasoning that a reviewer needs to validate the design.
  9. Save the final output to a markdown file named gh-pr.md.

Output format

Use this structure:

# <Short PR title>

<One short synopsis paragraph explaining the change and why it exists.>

- Closes <gh-issue-link> (If applicable, not all PRs pertain to issues)

## Rationale

<What was inconsistent, missing, or unsafe? Why is this event/data/model
boundary the right fix? Which related semantics deliberately do not change?>

## Key Changes

1. <Important change with rationale>
1. <Important change with rationale>

## 1 - <Behavior or logic topic>

```text
old source / behavior  -> replacement
authority source       -> fallback when unavailable
```

<Explain the important behavior, invariants, compatibility boundary, and
decision-making context. When useful, lead with a compact ASCII field map,
lifecycle, or state flow.>

## 2 - <Another behavior or logic topic>

<Add numbered topic-specific sections only when the PR has more than one
distinct behavior reviewers need to understand. The numbers aid scanning; they
do not describe implementation or runtime sequence.>

Style

  • Prefer clarity over cleverness.
  • Do not repeat the same point in multiple sections.
  • Lead with the outcome, then give reviewers the rationale and logic they need to assess correctness.
  • Explain non-obvious tradeoffs explicitly. Examples: a new field is canonical but a legacy fallback remains for historical data; a total intentionally excludes a separate accounting component; or an operation preserves a state-transition event while moving fee details to a normal receipt.
  • Always place ## Key Changes immediately after ## Rationale.
  • Use numbered lists for the ## Key Changes section, writing each item with 1. so the markdown source is easy to reorder.
  • Keep the description easy to scan with short paragraphs and direct headings. Prefer behavior-oriented sections over a chronological implementation diary. Keep Rationale and Key Changes unnumbered, and number only subsequent topic headings as ## 1 - ..., ## 2 - ..., and so forth. This numbering is for scanning, not sequencing; use only as many sections as the distinct fixes need.
  • For multi-source, lifecycle, migration, or fallback behavior, prefer compact ASCII diagrams and review invariants over prose-only explanations. Do not use Mermaid.
  • Exclude routine validation logs. Do not create dedicated validation or testing sections, and mention test strategy or coverage only when it is a meaningful change or reviewer risk.