agentsclimarketplace

Pr authoring

Skill Amey-Thakur/AI-SKILLS/skills/code-quality/pr-authoring

Shape pull requests a reviewer can actually hold in their head, through small size, a why-first description, and a self-review pass. Use when opening a pull request or preparing a change for review.From its SKILL.md

Install
npx -y skills add Amey-Thakur/AI-SKILLS --skill pr-authoring

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • 26 days oldThe repository was created 26 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
  • 4 stars4 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

2.4 KB, 512 tokens by cl100k_base, as published. Nobody here has run it

PR authoring

A pull request is a request for someone else's attention and judgment. A reviewer holds a small change well and skims a large one, so a 1000-line PR collects a rubber stamp exactly where it most needs scrutiny. Shape the PR so a careful reader can keep the whole change in their head.

Method

  1. Keep it under 400 lines of diff. Defect detection falls off a cliff past a few hundred changed lines. If the work is bigger, split it: a refactor PR, then the feature PR stacked on top. Two readable PRs beat one boulder.
  2. Lead the description with why, then what. First paragraph: the problem and the approach. Then the notable decisions and trade-offs. Link the issue for backstory rather than making the reviewer reconstruct intent from the diff.
  3. Do the first review yourself. Before requesting anyone, read your own diff top to bottom in the PR view. That is where you catch the stray console.log, the commented-out block, the file you did not mean to touch. Fix them so the reviewer spends on logic.
  4. Annotate the non-obvious inline. Leave author comments where a choice looks odd: "kept the loop instead of map because it short-circuits." Steer the reviewer past questions you can already answer.
  5. Separate mechanical commits from meaningful ones. A pure rename or format pass in its own commit lets the reviewer skim it and focus on the behavior commit. A mixed commit forces line-by-line reading of noise.
  6. Green the checks before requesting review. CI passing, tests added, lint clean. A red PR says "not ready" and burns a review cycle. Requesting review is a claim that you believe it is mergeable.

Checks

  • Is the diff small enough that a reviewer can read every line in one sitting?
  • Does the description answer why before what?
  • Did your own pass catch anything the reviewer should not have had to?

Boundaries

Large mechanical migrations (a codemod, a dependency bump touching many files) are legitimately big: flag them as such and point reviewers at the few files that need real eyes. Defer to the team's PR template and merge policy where one exists.

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.

Keep looking

Skills are one crate of 326,569. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.