Skip to main content
skill CX Codex

detailed-pr-description — Claude Code skill

Drafts GitHub PR descriptions with context, risks and validation results, or judges whether the branch has enough tests.

Install into Claude Code · Skills CLI (Vercel) All methods and platforms ↓
npx skills add biw/skills --skill detailed-pr-description -a claude-code
Open on GitHub

What detailed-pr-description does

The skill has three outcomes. Assessment answers whether the tests on a branch are strong enough and leaves the PR body alone. Draft builds a description locally or in the reply. Update and post write to GitHub, and only after you explicitly ask to update, edit, post or publish. Having gh on the command line is not permission to change a PR.

It starts by resolving the PR and its base through gh pr view --json number,title,body,baseRefName,headRefName,url, then gathers evidence from git status --short, diff stats, commits, changed files, tests, migrations and config. Drafts pull in assets/pr-body.md.tmpl unless the repository ships its own template. Every body needs one fenced excerpt of genuinely changed code, capped at 15 lines; a review section made of links will not do.

Tests get inventoried, and the skill runs the narrowest reliable checks first, then broader documented ones where feasible. Coverage lands in one of three buckets: sufficient, partially sufficient, insufficient. Confidence never comes from tests that were not run or do not apply. Unrelated code changes stay out unless you ask for them.

What it helps with

  • write a PR body with context, risks and follow-up work
  • decide whether branch tests are strong enough before review
  • record failed, skipped and unavailable checks with reasons
  • point reviewers at exact files and lines
  • refresh an existing PR description with gh pr edit

Example prompts for the agent

  • Assess whether the tests on this branch are sufficient, don't touch the PR body
  • Draft a PR description from the changes on this branch
  • Update the description of PR 42 on GitHub

How to install detailed-pr-description

1Skills CLI (Vercel)

npx skills add biw/skills --skill detailed-pr-description -a claude-code

2Manually: put the files in place

~/.claude/skills/detailed-pr-description/SKILL.md

Copy the whole skill folder; for one project use .claude/skills/detailed-pr-description/

Step by step, with commands for each platform: How to install skills in Claude Code: five ways

Questions about detailed-pr-description

How do I install detailed-pr-description in Claude Code?

Run npx skills add biw/skills --skill detailed-pr-description -a claude-code. The skill lands at ~/.claude/skills/detailed-pr-description/SKILL.md.

What does detailed-pr-description do?

It either assesses test coverage on a branch or writes and updates a GitHub PR description covering context, risks, follow-up work, code references and validation results.

Does the skill edit my PR on GitHub on its own?

Only when you ask it to update, edit, post or publish. In draft mode it writes .context/pr-description.md and never calls gh pr edit.

Source text

SKILL.md · MIT
---
name: detailed-pr-description
description: Assess test coverage or draft/update GitHub PR descriptions with self-contained context, risks, follow-up work, focused code references, and validation results.
---

# Detailed PR Description

## Resolve the outcome

- **Assessment:** judge whether branch tests are strong enough; do not draft or update a PR body.
- **Draft:** produce a proposed body locally or in the response; do not mutate GitHub.
- **Update/post:** draft and update GitHub only when explicitly asked to update, edit, post, or publish.

Perform each requested outcome when combined. CLI access is never authorization to mutate a PR.

## Workflow

1. Identify the PR and base. Prefer `gh pr view --json number,title,body,baseRefName,headRefName,url`; without a PR, compare against the user-provided or repository default base. Fetch before relying on stale refs.
2. Establish evidence from `git status --short`, diff stats and names, relevant commits, changed files, tests, docs, migrations, config, and generated artifacts. Treat `.context/` as scratch and restate any useful fact because it is untracked. Do not rely on chat or memory for behavioral claims.
3. Determine the problem, approach, outcome, deliberate exclusions, risk, compatibility, edge cases, operational gotchas, migrations, flags, cleanup, and reviewer focus. For close-review items, give a precise file/line reference and explanation. Every PR body must include one fenced excerpt of actual changed code, configuration, or documentation from the most review-critical hunk, capped at 15 lines; a link-only review section is insufficient.
4. Inventory affected and added tests. Run the narrowest reliable checks, then documented broader validation when feasible. Classify coverage as sufficient, partially sufficient, or insufficient; name regressions it catches, important gaps, and concrete follow-up tests. Never claim confidence from unrun or irrelevant tests.
5. Deliver only the authorized outcome:
   - Assessment: report evidence and coverage judgment without creating a body.
   - Draft: return the body or write `.context/pr-description.md`; never call `gh pr edit`.
   - Update/post: write `.context/pr-description.md`, then use `gh pr edit <number> --body-file .context/pr-description.md`. If posting fails, preserve the draft and explain why.

## PR body

For draft/update work, load `assets/pr-body.md.tmpl` unless the repository has a stronger template. The body must stand alone without chat or local notes. Prefer high-signal context; add screenshots, rollout, migrations, flags, performance, accessibility, security, or release notes only when relevant. Remove an empty section only when it genuinely does not apply; use `Not applicable` when omitting it could hide an important review dimension.

## Quality bar

- Ground claims in code, tests, docs, commits, or command output; distinguish facts from assumptions and judgment.
- Preserve the existing title and useful body content unless replacement is clearer.
- Name behavior instead of saying “various fixes” or “improves behavior.”
- Cite precise file/line references for close-review items. Include one actual changed excerpt in a fenced code block; it must be materially helpful and at most 15 lines.
- Include failed, skipped, and unavailable validation with reasons.
- Do not make unrelated code changes unless explicitly asked.

## Final response

Report the PR URL when one exists, whether the outcome was an assessment, local draft, or posted update, validation commands/results, and the bottom-line coverage assessment. If an authorized post failed, report the draft path.

© biw · MIT · biw/skills

Similar skills

All skills for Git and GitHub →

From the same set3

Project sponsors

Companies that keep this analytics open
Gold sponsor

Impressions, clicks and code copies — open statistics. We keep the numbers public for the same reason NeBlask itself is public. Become a sponsor