Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →naxodev/pr-walkthrough
Generates a narrative HTML walkthrough of a pull request, with optional embedded audio narration — an essay covering problem, motivation, trade-offs, alternatives, and file tour.
Passed 11 of 18 test cases with the skill, and 5 of 18 without it.
and 6 more in the full report
9 fixed5 passed either way7 still failing2 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 9 test cases the skill fixed. Same model, same prompt, with and without the skill.
What top-level JSON field in the walkthrough manifest holds the audio opening spoken prior to section 1?
| Checked | Without skill | With skill |
|---|---|---|
| Names narrationIntro as the manifest field for the spoken opening. | Missed | Passed |
The instructions your model reads when the skill loads.
Produces an essay-shaped HTML walkthrough of a pull request. Output is a story, not a review: it explains where the system was, what gap or motivation prompted the change, and what was built, with documented limitations and follow-ups where relevant.
Read STYLE.md before writing. It defines the voice, required arc, and when optional sections earn their place.
Steps in order. Steps 2 and 3 are what keep the output grounded — run them every time.
Relative paths below resolve from the skill root, the directory containing this SKILL.md.
Take the PR number from the user's argument. If none given, look it up for the current branch:
gh pr view --json number --jq .number
If that fails (no PR for current branch), ask the user.
Complete when: the PR number is known; for a small PR, the user has chosen the full walkthrough (including an earlier insistence) or the workflow stops for the requested short summary.
node scripts/fetch-pr.mjs <PR#>
Prints the path to a JSON dump in /tmp containing PR metadata + full diff. Use that path as input for all subagents below.
Complete when: the fetch command succeeded and returned a readable JSON dump path.
Dispatch all four subagents in parallel — in Claude Code, four Agent tool calls (type general-purpose) in a single message; on other agents, the native subagent/task mechanism. Pass the JSON dump path so each agent can read it instead of being handed a giant inline diff.
| Agent | Charter |
|---|
test cases passed, vs 5 / 18 without
Published skills whose descriptions closely match this one, with their headline benchmark lift.
More skills from the same source and nearby in the registry, with their headline benchmark lift.