Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Review documentation changes for compliance with the Metabase writing style guide. Use when reviewing pull requests, files, or diffs containing documentation markdown files.
.claude/skills/microck-docs-review/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-06 | ✗→✓ | ▲ Improved | 134% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 100% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 109% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 55% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 92% | 0% |
@./../_shared/metabase-style-guide.md
IMPORTANT: Before starting the review, determine which mode to use:
mcp__github__create_pending_pull_request_review tool is available, you are reviewing a GitHub PRmcp__github__create_pending_pull_request_review is availableRun through the diff looking for these issues:
Tone and voice:
Structure and clarity:
Links and references:
Formatting:
Code and examples:
Sentence construction:
| Pattern | Issue | | ----------------------------- | --------------------------------------------- | | Button name or UI element | Should use bold not backticks | | we can do X, our feature | Should be "Metabase" or "it" | | click here, read more here | Need descriptive link text | | easy, simple, just | Remove condescending qualifiers | | users | Should be "people" or "companies" if possible |
MANDATORY REQUIREMENT: Every single issue MUST be numbered sequentially starting from Issue 1.
This numbered format is NON-NEGOTIABLE. It allows users to efficiently reference specific issues (e.g., "fix issues 1, 3, and 5") and track which feedback has been addressed.
When outputting issues in the conversation (local mode), use this format:
markdown## Issues **Issue 1: [Brief title]** Line X: Succinct description of the issue [code or example] Suggested fix or succinct explanation **Issue 2: [Brief title]** Line Y: Description of the issue Suggested fix or explanation **Issue 3: [Brief title]** ...
Examples:
> Issue 1: Backticks on UI elements > Line 42: This uses backticks for the UI element. Use bold instead: Filter not Filter.
> Issue 2: Formal tone > Line 15: This could be more conversational. Consider: "You can't..." instead of "You cannot..."
> Issue 3: Vague heading > Line 8: The heading could be more specific. Try stating the point directly: "Run migrations before upgrading" vs "Upgrade process"
When posting to GitHub (PR mode), use the pending review workflow:
Workflow steps:
mcp__github__create_pending_pull_request_review to begin a pending reviewmcp__github__get_pull_request_diff to understand the code changes and line numbersmcp__github__add_pull_request_review_comment_to_pending_review for each issue**Issue N: [Brief title]**mcp__github__submit_pending_pull_request_review to publish all comments at once"COMMENT" (NOT "REQUEST_CHANGES") to make it non-blockingComment format example:
**Issue 1: Backticks on UI elements**
This uses backticks for the UI element. Use **bold** instead: **Filter** not `Filter`.IMPORTANT:
**Issue N: [Brief title]****Issue N: [Brief title]** where N is the issue number.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 3,666 | 9,625 | +163% | 1 | 1 | 0% | 483 | 2,144 | +344% | 0 | 0 | — |
case-02 | pass→pass | 15,438 | 15,004 | -3% | 1 | 1 | 0% | 2,692 | 4,568 | +70% | 0 | 0 | — |
case-03 | pass→pass | 8,140 | 6,791 | -17% | 1 | 1 | 0% | 1,255 | 2,606 | +108% | 0 | 0 | — |
case-04 | pass→pass | 7,199 | 6,771 | -6% | 1 | 1 | 0% | 1,342 | 2,848 | +112% | 0 | 0 | — |
case-05 | pass→pass | 7,614 | 5,225 | -31% | 1 | 1 | 0% | 1,290 | 2,398 | +86% | 0 | 0 | — |
case-06 | fail→pass | 7,222 | 6,425 | -11% | 1 | 1 | 0% | 1,159 | 2,707 | +134% | 0 | 0 | — |
case-07 | fail→pass | 8,597 | 6,003 | -30% | 1 | 1 | 0% | 1,313 | 2,627 | +100% | 0 | 0 | — |
case-08 | pass→pass | 7,307 | 3,972 | -46% | 1 | 1 | 0% | 1,154 | 2,165 | +88% | 0 | 0 | — |
case-09 | fail→pass | 7,437 | 5,328 | -28% | 1 | 1 | 0% | 1,205 | 2,516 | +109% | 0 | 0 | — |
case-10 | pass→pass | 10,575 | 3,054 | -71% | 1 | 1 | 0% | 1,847 | 2,141 | +16% | 0 | 0 | — |
case-11 | fail→pass | 8,066 | 2,356 | -71% | 1 | 1 | 0% | 1,265 | 1,962 | +55% | 0 | 0 | — |
case-12 | fail→pass | 6,804 | 9,352 | +37% | 1 | 1 | 0% | 1,124 | 2,161 | +92% | 0 | 0 | — |
case-13 | fail→pass | 11,172 | 7,826 | -30% | 1 | 1 | 0% | 1,744 | 2,746 | +57% | 0 | 0 | — |
case-14 | pass→pass | 7,937 | 7,270 | -8% | 1 | 1 | 0% | 1,311 | 2,793 | +113% | 0 | 0 | — |
case-15 | pass→pass | 6,303 | 5,038 | -20% | 1 | 1 | 0% | 1,044 | 2,443 | +134% | 0 | 0 | — |
case-16 | pass→pass | 6,699 | 7,089 | +6% | 1 | 1 | 0% | 1,218 | 2,658 | +118% | 0 | 0 | — |
case-17 | pass→pass | 13,062 | 5,644 | -57% | 1 | 1 | 0% | 2,602 | 2,579 | -1% | 0 | 0 | — |
case-18 | pass→pass | 9,412 | 2,484 | -74% | 1 | 1 | 0% | 1,428 | 1,948 | +36% | 0 | 0 | — |
case-19 | fail→pass | 14,899 | 3,828 | -74% | 1 | 1 | 0% | 2,519 | 2,206 | -12% | 0 | 0 | — |
case-20 | pass→pass | 6,567 | 2,224 | -66% | 1 | 1 | 0% | 999 | 1,970 | +97% | 0 | 0 | — |
case-21 | pass→pass | 8,203 | 5,491 | -33% | 1 | 1 | 0% | 1,139 | 2,527 | +122% | 0 | 0 | — |
case-22 | pass→pass | 6,836 | 5,425 | -21% | 1 | 1 | 0% | 1,062 | 2,442 | +130% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 21 counted toward the lift figure. The other 1 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +32 percentage points is the difference between those two pass rates over the 21 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.