| name | pr-check |
| description | Validate a PR against PR standards before requesting review. Checks branch naming, title, description, commits, references, and scope. Triggers: /pr-check, check this PR, validate PR |
| license | MIT |
| metadata | {"version":"2.0.0","tags":["pr","check","standards","validation"],"author":"benjamcalvin"} |
PR Standards Check
Suite capability: focused-only. Run focused tests, lint, builds, and acceptance commands only. Do not execute or consume the target repository's authoritative verification command or ordered command plan. Final verification owns that evidence.
Pre-flight validation for PRs.
$ARGUMENTS
If the current client leaves $ARGUMENTS literal, use the user's invoking prompt. If no PR number is supplied, inspect the PR associated with the current branch when available.
Context
At runtime, inspect the current branch and fetch available PR metadata and comments. Determine the actual base branch before inspecting commits and diff size. Read explicit user direction and the target repository's governing instructions and contribution documentation before applying standards.
Instructions
Validate the current PR against target-repository policy first. Explicit user direction and repository instructions take precedence over the bundled checks below for branch names, commits, PR formatting, required references, and blocking/advisory status. Use a bundled check only as a fallback where target policy is silent, and identify that fallback in the result. If no PR exists, check only what can be validated locally and note that no PR exists yet.
For each check, output one of:
- PASS — Meets the standard
- WARN — Minor deviation, note what's off
- FAIL — Does not meet the standard, explain what needs to change
Checks
1. Branch Naming (bundled fallback)
Branch must match <type>/<short-description> where type is one of: feat, fix, refactor, docs, test, chore. Must be lowercase, hyphen-separated.
2. PR Title (bundled fallback)
Must match <type>: <imperative summary>. Type prefix should match branch type. Under 72 characters. No period at the end. Imperative mood ("Add", "Fix"), not past tense ("Added", "Fixed").
3. PR Description — TL;DR (bundled fallback)
Must open with 2-4 plain-language sentences explaining what the change does and why (a ## TL;DR or ## Summary section, before any mechanism). FAIL if the description opens with implementation bullets instead of prose a non-reader of the code could follow.
4. PR Description — Test Evidence (bundled fallback)
Must include how the change was verified: test output, manual steps, or "covered by existing tests."
5. Commit Messages (bundled fallback)
Each commit message should follow <type>: <summary> format. No "WIP", "fixup", or "wip" commits.
6. References (bundled fallback)
If the change relates to a GitHub issue, it should reference it with an appropriate keyword:
Closes #N / Fixes #N — only when this single PR fully completes the issuePart of #N — when the PR is one of several addressing the issue
WARN if no references found (not all PRs need them, but flag for awareness). WARN if Closes #N is used but the PR appears to be a sub-task of a larger issue (e.g., the issue has multiple acceptance criteria and the PR only addresses some).
7. Altitude Layering (bundled fallback)
The description must descend through altitude layers rather than mixing them:
- The TL;DR contains no file paths, function names, or line numbers — behavior in plain language only
- Design reasoning (when present) is in component terms; code identifiers appear only in implementation-level sections (Implementation Notes, Test evidence, Review focus)
- No sentence carries a claim, its mechanism, and a citation at once; citations sit at the end of their bullet, not mid-clause
- No review chronology woven into the description ("round 1 added...", "after feedback we...")
WARN on isolated violations; FAIL if the TL;DR is saturated with code identifiers or the layers are absent entirely.
Scope Note (advisory — not scored)
Assess whether the PR is one logical change a reviewer can hold in their head in one sitting, or a small batch of same-kind housekeeping changes. This is a judgment about cohesion, not size: a single ADR, a new module's scaffold, and a mechanical rename are each one logical change however many lines they span, while a small PR that fixes a bug and refactors an unrelated module is two.
Report a one-line observation. Say the PR is cohesive, or name the seam it should be split along. Do not assign PASS/WARN/FAIL, do not count lines against a threshold, and do not treat this note as a merge blocker — it exists to inform the author and reviewers, not to gate.
Size-to-cap advisory (early split signal). If the PR is over-scoped — far larger than the change it claims to be, or spanning multiple unrelated changes a reviewer cannot hold in one sitting — flag it and recommend splitting the out-of-scope work into child issues BEFORE implementation proceeds. This is a qualitative scope judgment, not a line-count threshold (see the scope note above). A PR that balloons past reviewable size mid-lifecycle forces scope-splitting refine passes and extra review rounds — far cheaper to split up front. This is advisory, not a blocker.
Output Format
## PR Standards Check
| # | Check | Result | Notes |
|---|-------|--------|-------|
| 1 | Branch naming | PASS/WARN/FAIL | ... |
| 2 | PR title | PASS/WARN/FAIL | ... |
| 3 | TL;DR | PASS/WARN/FAIL | ... |
| 4 | Test evidence | PASS/WARN/FAIL | ... |
| 5 | Commit messages | PASS/WARN/FAIL | ... |
| 6 | References | PASS/WARN/FAIL | ... |
| 7 | Altitude layering | PASS/WARN/FAIL | ... |
**Result: X/7 passing, Y warnings, Z failures**
**Scope (advisory):** <one line — cohesive, or the seam it should be split along>
If there are failures, add a brief "Suggested Fixes" section listing what to change.
