Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Implement CodeRabbit reference architecture with production-grade .coderabbit.yaml configuration. Use when designing review configuration for a new project, establishing team standards, or building a comprehensive review setup from scratch. Trigger with phrases like "coderabbit architecture", "coderabbit best practices", "coderabbit project structure", "coderabbit reference config", "coderabbit full setup".
.claude/skills/jeremylongshore-coderabbit-reference-architecture/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 1% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 43% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 100% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 92% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 37% | 0% |
Make authority and data flow explicit. Separate source, control planes, provider events, knowledge sources, and reporting.
references/official-docs.md and re-check any time-sensitive contract before execution.Treat Git-provider sessions, CodeRabbit web sessions, CLI credentials, and CodeRabbit API keys as separate credentials. Use only an already-approved session or secret-manager reference, never print a secret, and do not place credentials in .coderabbit.yaml, source files, logs, or deliverables.
Require architecture, security, privacy, repository, and platform owners before scope expansion. Keep analysis and drafts local until approval is explicit, and record who approved the action and its scope.
A boundary diagram, authority matrix, data flows, controls, failure modes, and phased plan. Include source dates, unknowns, and the exact boundary between observed fact and recommendation.
| Condition | Response | |---|---| | Current contract is unclear or docs disagree | Stop mutation, cite both sources, and request owner resolution. | | Required access or approval is missing | Produce a draft and evidence plan only. | | Validation or pilot behavior differs from expectation | Restore the prior state and retain the failed evidence. | | Output contains secrets or private code | Stop, quarantine the artifact, redact it, and notify the data owner. |
Design central defaults with local exceptions and global security overrides.
Add MCP context only after data scope and auth approval.
references/official-docs.md.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-18 | pass→pass | 7,005 | 2,075 | -70% | 1 | 1 | 0% | 1,126 | 2,703 | +140% | 0 | 0 | — |
case-01 | fail→pass | 33,843 | 18,527 | -45% | 1 | 1 | 0% | 6,203 | 6,257 | +1% | 0 | 0 | — |
case-02 | pass→pass | 24,861 | 21,258 | -14% | 1 | 1 | 0% | 4,093 | 6,487 | +58% | 0 | 0 | — |
case-03 | fail→pass | 26,955 | 24,408 | -9% | 1 | 1 | 0% | 4,917 | 7,043 | +43% | 0 | 0 | — |
case-04 | fail→fail | 8,956 | 4,966 | -45% | 1 | 1 | 0% | 1,516 | 3,304 | +118% | 0 | 0 | — |
case-05 | fail→pass | 9,155 | 5,563 | -39% | 1 | 1 | 0% | 1,770 | 3,537 | +100% | 0 | 0 | — |
case-06 | fail→pass | 9,740 | 6,443 | -34% | 1 | 1 | 0% | 1,920 | 3,681 | +92% | 0 | 0 | — |
case-07 | pass→fail | 14,477 | 8,211 | -43% | 1 | 1 | 0% | 2,528 | 3,916 | +55% | 0 | 0 | — |
case-08 | pass→pass | 12,064 | 11,899 | -1% | 1 | 1 | 0% | 1,944 | 4,354 | +124% | 0 | 0 | — |
case-09 | pass→pass | 10,315 | 9,993 | -3% | 1 | 1 | 0% | 1,879 | 4,184 | +123% | 0 | 0 | — |
case-10 | fail→pass | 11,985 | 3,097 | -74% | 1 | 1 | 0% | 2,206 | 3,021 | +37% | 0 | 0 | — |
case-11 | fail→pass | 13,362 | 12,422 | -7% | 1 | 1 | 0% | 2,555 | 4,784 | +87% | 0 | 0 | — |
case-12 | pass→pass | 12,551 | 9,720 | -23% | 1 | 1 | 0% | 2,235 | 4,334 | +94% | 0 | 0 | — |
case-13 | fail→pass | 13,631 | 11,405 | -16% | 1 | 1 | 0% | 2,483 | 4,759 | +92% | 0 | 0 | — |
case-19 | pass→pass | 6,521 | 2,180 | -67% | 1 | 1 | 0% | 1,196 | 2,864 | +139% | 0 | 0 | — |
case-14 | fail→pass | 14,316 | 5,014 | -65% | 1 | 1 | 0% | 2,872 | 3,402 | +18% | 0 | 0 | — |
case-15 | fail→pass | 13,487 | 6,965 | -48% | 1 | 1 | 0% | 2,634 | 3,704 | +41% | 0 | 0 | — |
case-16 | pass→pass | 9,576 | 2,752 | -71% | 1 | 1 | 0% | 1,766 | 2,867 | +62% | 0 | 0 | — |
case-17 | pass→pass | 11,991 | 6,215 | -48% | 1 | 1 | 0% | 2,352 | 3,700 | +57% | 0 | 0 | — |
case-20 | pass→pass | 11,883 | 4,915 | -59% | 1 | 1 | 0% | 2,046 | 3,345 | +63% | 0 | 0 | — |
case-21 | pass→pass | 6,195 | 3,706 | -40% | 1 | 1 | 0% | 1,038 | 3,202 | +208% | 0 | 0 | — |
case-22 | pass→pass | 3,471 | 2,264 | -35% | 1 | 1 | 0% | 599 | 2,835 | +373% | 0 | 0 | — |
case-23 | pass→pass | 3,662 | 4,667 | +27% | 1 | 1 | 0% | 683 | 3,349 | +390% | 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. 23 cases were attempted. The headline lift of +35 percentage points is the difference between those two pass rates over the 23 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
The publisher has shipped newer versions since this run, so these numbers describe v1, not the version currently listed.
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.