Install
Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.
npx agentmarket install skill/michielvandermeer-pr
Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Use cases: writing
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions
What this skill can do and what it asks for, from its SKILL.md and code. Checked Oct 25, 2026.
- Tools
- No tools found
- Secrets
- None asked for
- Install scripts
- None found
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/pr
What's inside
- Skill name
pr- Description
- Pull request descriptions. Use when writing the description of a pull request or merge request.
- License
- MIT
- Instructions
- 72 lines in SKILL.md · view on GitHub
Files (1)
- SKILL.md
SKILL.md (first lines)
# Pull request description
A description shows a reviewer what changed, that it works, and how risky it is to merge, in three sections, in this order:
```markdown
## What changes
<the smallest picture that makes the change clear, with the sentence it supports>
## Evidence
<the one fact the change is safe because of, naming what breaks if it is false>
**Before:** <output, screenshot, or one sentence>
**After:** <output or screenshot>
## Merge risk
**Undo:** <easy. Reverting the commits restores the old behaviour. | hard. What stays changed after a revert.>
**Affects:** <who or what would notice if it went wrong>
<the Issue's closing reference>
```
Its reader is a reviewer who was never in the session. Read [`technical-writing/SKILL.md`](../technical-writing/SKILL.md) and write it at that bar (`/technical-writing` does not load as a skill): the repo's `CONTEXT.md` terms stand as they are, and this plugin's own words — Step, Iteration, Verifier, run folder — stay out. When the work came from an Issue, end with the closing reference the repo's Tracker ref (`.agents/refs/tracker.md`) gives; a ref that gives none gets the Issue's reference on its own line.
Write the description to a file outside the repo, in a folder made with `mktemp -d` that no other session shares, and pass it with `--body-file` to `gh pr create` or `gh pr edit`, or the host's equivalent.
## What changes
Pick the smallest view that makes the point. One is usual and several is fine:
- Logic or an algorithm: pseudocode.
- Runtime control flow: a call tree.
- UI structure: a component tree, with the state and module boundaries that matter.
- File responsibility or a broad refactor: a shallow file tree, one comment per folder.
- Interaction or data flow between parts: a Mermaid sequence diagram.
- A change inside a shape that already exists: a `diff` of that shape — component tree, file tree, call tree, or control flow — rather than of the code:
```diff
submitForm
createSession
persistPrompt
+ expandSkillMention
launchAgent
```
- Code that is mostly new, or whose order matters: the whole block.
Keep only the calls, files, props, and states the point needs, and put each picture next to the sentence it supports.
## Evidence
Open with the one fact the change is safe because of, naming what breaks if it is false. Then show the check that proves it:
- **After**: output from a run on the exact code being pushed. When the code changed since the check last ran, run it again.
- **Before**: the same check on the target branch, in a throwaway worktree, when it can fail there. Otherwise one sentence on what happened before.
A visual change shows a screenshot, or a short recording when the change is an interaction, taken by driving the app the way the repo's Verify skill (`.claude/skills/verify/`) says when it has one. Save it outside the repo, reference it in the description by that path, and upload it with `--attach '<path>#<alt text>'`, which swaps the path for the uploaded link (`gh` 2.99 and later, on GitHub.com). Where the host's tool cannot attach — GitHub Enterprise Server, GitLab, an older `gh` — leave `<!-- add <path> here -->` in its place and tell the user which file to add.
A change to no code that runs, such as a docs edit, says in one line that it needs no evidence. A check you could not run is named with the reason, in place of its output: every line of output in the description comes from a run you saw.
After an implement run in this session, its final report names the Hand checks the Verifier ran on the code that landed. Use them as the after half unless the code changed since. The safety fact and Merge risk are yours to work out from the diff.
## Merge risk
**Undo** is easy when reverting the commits restores the old behaviour, and hard when something stays changed after the revert — data written, a migration run, a file format or API others already read, a release published. A hard Undo names what stays changed.
**Affects** names who or what would notice if the change went wrong: the users of a screen, the callers of an API, a nightly job, the layout on a phone.From michielvandermeer/skills, MIT.
More from michielvandermeer/skills
- 0.628score
implement
michielvandermeer
Implement a spec by slicing it into steps, then a loop of implement, review, and verify, each in its own sub-agent.
SkillAFreshness: A★ 9Claude Code +9 - 0.628score
technical-writing
michielvandermeer
The house standard for every sentence a person reads.
SkillAFreshness: A★ 9Claude Code +9 - 0.628score
verify
michielvandermeer
Drive this checkout's skills the way a user does — load the working tree as the plugin in a headless Claude Code session inside a scratch repo, type the slash…
SkillAFreshness: A★ 9Claude Code +9 - 0.628score
hillclimb
michielvandermeer
Hillclimb one measured number — test-suite time, build time, memory, bundle size — keeping only the changes that measurably help.
SkillAFreshness: A★ 9Claude Code +9 - 0.625score
doctor
michielvandermeer
Moves documents into this project's canonical layout, closes local Issues the code already implements, and brings every ADR to state the decision in force.
SkillAFreshness: A★ 9Claude Code +9 - 0.625score
explain
michielvandermeer
Explain anything as a written walkthrough, a diagram in the chat, an HTML page, or a narrated video, picking the format that fits.
SkillAFreshness: A★ 9Claude Code +9 - 0.625score
retro
michielvandermeer
Look at a finished session, apply high-priority changes this repository owns, file the structural ones as Issues, and summarise the rest.
SkillAFreshness: A★ 9Claude Code +9 - 0.625score
setup
michielvandermeer
Set up or update the repo's Tracker ref, where its Issues live; its Verify skill, which drives the app the way a user does; or its Models ref, the model and…
SkillAFreshness: A★ 9Claude Code +9 - 0.622score
learn-habits
michielvandermeer
Rebuild a person's Grilling profile from the Habits in their past grilling sessions.
SkillAFreshness: A★ 9Claude Code +9 - 0.622score
brainstorm
michielvandermeer
Explore a vague problem as very different Directions, side by side, and file the ones you keep as Issues.
SkillAFreshness: A★ 9Claude Code +9 - 0.622score
triage
michielvandermeer
Sort incoming reports into Specs and parked Issues — one Issue per distinct problem.
SkillAFreshness: A★ 9Claude Code +9 - 0.620score
grill-with-docs
michielvandermeer
A relentless round-by-round interview to sharpen a plan or design, which also creates docs (ADRs and glossary) as we go.
SkillAFreshness: A★ 9Claude Code +9
Signals
- Stars
- 9
- Forks
- 0
- Open issues
- 8
- Last commit (this folder)
- Oct 17, 2026 (12 days ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- MIT
- Repository
- github.com/michielvandermeer/skills
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from michielvandermeer
- 0.628score
implement
michielvandermeer
Implement a spec by slicing it into steps, then a loop of implement, review, and verify, each in its own sub-agent.
SkillAFreshness: A★ 9Claude Code +9 - 0.628score
verify
michielvandermeer
Drive this checkout's skills the way a user does — load the working tree as the plugin in a headless Claude Code session inside a scratch repo, type the slash…
SkillAFreshness: A★ 9Claude Code +9 - 0.628score
technical-writing
michielvandermeer
The house standard for every sentence a person reads.
SkillAFreshness: A★ 9Claude Code +9 - 0.628score
hillclimb
michielvandermeer
Hillclimb one measured number — test-suite time, build time, memory, bundle size — keeping only the changes that measurably help.
SkillAFreshness: A★ 9Claude Code +9 - 0.625score
explain
michielvandermeer
Explain anything as a written walkthrough, a diagram in the chat, an HTML page, or a narrated video, picking the format that fits.
SkillAFreshness: A★ 9Claude Code +9 - 0.625score
doctor
michielvandermeer
Moves documents into this project's canonical layout, closes local Issues the code already implements, and brings every ADR to state the decision in force.
SkillAFreshness: A★ 9Claude Code +9
Cite this page
pr skill. Agent Mart. https://agentmart.dev/skills/michielvandermeer-pr (last updated 2026-10-24)