Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →mlflow/pr-review
Review a pull request and emit a validated review payload.
Passed 12 of 18 test cases with the skill, and 6 of 18 without it.
and 14 more in the full report
11 fixed6 passed either way5 still failing3 broke
Tokens, turns and time describe resource use, not a dollar price.
This benchmark measured an earlier version of the skill.
One of 10 test cases the skill fixed. Same model, same prompt, with and without the skill.
You are beginning the analysis stage of an MLflow PR review at checkout directory /repo/checkout. You want to check repository-specific coding standards for the modified files. Should you manually read all .md files in the repository's docs folder, or is there a specific CLI command to load rule definitions for changed files?
| Checked | Without skill | With skill |
|---|---|---|
| Passes changed file names from git diff into skills load-rules. | Missed | Passed |
| Includes the --directory flag pointing to the base working directory for uv run. | Missed | Passed |
The instructions your model reads when the skill loads.
Review $pr_url and write a JSON review payload to $payload_path. Do not post anything: writing that payload is the whole job.
The PR is checked out at $pr_checkout, not in the working directory:
$base_dir # the working directory: this skill and the `skills`
# CLI come from here, and nothing you review does
$pr_checkout # the reviewed tree: the PR merged into its base
The working directory holds a checkout of the same repository, so it looks like the code under
review and is not guaranteed to match it. Everything aimed at the PR needs the prefix:
git -C $pr_checkout ..., $pr_checkout/<path> to open or grep a file, and
cd $pr_checkout && ... for anything that builds or runs repository code. The
uv run commands below are the exception: --directory $base_dir pins each one to
this checkout, so it keeps using this tree's skills package and rules even when the
working directory has moved into $pr_checkout. uv resolves its workspace from the
working directory, and so does the rule loader, so dropping the flag silently hands
both to the code under review.
The commands below take <pr_url> from the PR URL above, and <owner>, <repo>, and
<pr_number> from its parts.
These reads are independent. Issue them as parallel tool calls in a single turn, not sequentially.
PR title and description
gh pr view <pr_url> --json title,body
test cases passed, vs 6 / 18 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.