Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Run a sequential three-tier code review on a finished implementation branch — Haiku (surface) → Sonnet (logic) → Opus (deep). Restarts from Tier 1 on any tier failure. Use when a solo branch or PR is code-complete and you want structured pre-merge verification before human review.
.claude/skills/davepoon-ralph-review-trio/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 126% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -40% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 12% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -29% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 20% | 0% |
This skill triggers /ralph-review, which runs three sequential reviewer subagents at increasing depth. If any tier flags a failure, the loop restarts from Tier 1 after fixes.
/ralph-review
│
▼
Tier 1 — Haiku (surface checks) ─── fail ──> restart
│ pass
▼
Tier 2 — Sonnet (logic checks) ─── fail ──> restart
│ pass
▼
Tier 3 — Opus (deep analysis) ─── fail ──> restart
│ pass
▼
RALPH_PASS → merge OKA <promise>HAIKU_PASS</promise> / <promise>SONNET_PASS</promise> / <promise>OPUS_PASS</promise> token is emitted by each tier on pass. All three required for overall pass.
/ralph-review — defined in ../../commands/ralph-review.md.
../../agents/haiku-reviewer.md — Tier 1 surface checklist../../agents/sonnet-reviewer.md — Tier 2 logic checklist../../agents/opus-reviewer.md — Tier 3 deep-analysis checklistExtended reference content under references/ (same dir as this SKILL.md) is loaded on demand by each tier when a specific check requires more context.
Each tier writes a fenced ## RESULT block with:
## RESULT
mcp_graph_available: yes|no # first line when discussing graph queries
verdict: pass|fail|unknown
files_touched: [paths]
findings: [{path, line, claim, evidence}]
scope_gaps: [list or "none"]The main agent reads the RESULT block and decides next action (restart, next tier, or PASS).
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 5,558 | 7,449 | +34% | 1 | 1 | 0% | 948 | 2,146 | +126% | 0 | 0 | — |
case-02 | fail→fail | 7,637 | 5,181 | -32% | 1 | 1 | 0% | 1,246 | 969 | -22% | 0 | 0 | — |
case-03 | fail→fail | 8,033 | 3,478 | -57% | 1 | 1 | 0% | 1,348 | 791 | -41% | 0 | 0 | — |
case-09 | fail→pass | 7,189 | 1,612 | -78% | 1 | 1 | 0% | 1,412 | 848 | -40% | 0 | 0 | — |
case-04 | pass→pass | 4,081 | 2,645 | -35% | 1 | 1 | 0% | 704 | 1,078 | +53% | 0 | 0 | — |
case-05 | fail→pass | 16,444 | 1,771 | -89% | 1 | 1 | 0% | 771 | 867 | +12% | 0 | 0 | — |
case-06 | fail→pass | 21,233 | 1,954 | -91% | 1 | 1 | 0% | 1,290 | 912 | -29% | 0 | 0 | — |
case-07 | fail→pass | 12,670 | 1,907 | -85% | 1 | 1 | 0% | 760 | 914 | +20% | 0 | 0 | — |
case-08 | fail→fail | 6,419 | 1,956 | -70% | 1 | 1 | 0% | 1,105 | 991 | -10% | 0 | 0 | — |
case-10 | fail→pass | 8,759 | 3,425 | -61% | 1 | 1 | 0% | 1,851 | 1,156 | -38% | 0 | 0 | — |
case-11 | pass→pass | 5,526 | 1,489 | -73% | 1 | 1 | 0% | 939 | 819 | -13% | 0 | 0 | — |
case-12 | pass→pass | 8,233 | 1,551 | -81% | 1 | 1 | 0% | 1,441 | 807 | -44% | 0 | 0 | — |
case-13 | fail→pass | 10,776 | 1,507 | -86% | 1 | 1 | 0% | 1,880 | 820 | -56% | 0 | 0 | — |
case-14 | fail→pass | 9,029 | 3,102 | -66% | 1 | 1 | 0% | 1,401 | 1,109 | -21% | 0 | 0 | — |
case-15 | fail→pass | 10,911 | 1,540 | -86% | 1 | 1 | 0% | 1,839 | 785 | -57% | 0 | 0 | — |
case-16 | fail→pass | 6,584 | 1,690 | -74% | 1 | 1 | 0% | 1,006 | 847 | -16% | 0 | 0 | — |
case-17 | fail→pass | 3,523 | 1,694 | -52% | 1 | 1 | 0% | 539 | 838 | +55% | 0 | 0 | — |
case-18 | fail→pass | 9,668 | 1,551 | -84% | 1 | 1 | 0% | 1,615 | 826 | -49% | 0 | 0 | — |
case-19 | pass→pass | 4,685 | 2,112 | -55% | 1 | 1 | 0% | 733 | 904 | +23% | 0 | 0 | — |
case-20 | pass→pass | 8,680 | 3,357 | -61% | 1 | 1 | 0% | 1,359 | 1,084 | -20% | 0 | 0 | — |
case-21 | pass→pass | 9,243 | 3,068 | -67% | 1 | 1 | 0% | 1,333 | 1,054 | -21% | 0 | 0 | — |
case-22 | pass→pass | 8,056 | 2,466 | -69% | 1 | 1 | 0% | 1,293 | 970 | -25% | 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 17 counted toward the lift figure. The other 5 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 +55 percentage points is the difference between those two pass rates over the 17 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.