Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Review an implementation plan for repo accuracy, fact purity, intent fidelity, reconciliation quality, and completeness. Use when a plan needs a correctness and completeness pass.
.claude/skills/dcouple-plan-reviewer/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-07 | ✗→✓ | ▲ Improved | 72% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 7% | 0% |
| case-16 | ✗→✓ | ▲ Improved | 102% | 0% |
| case-04 | ✓→✗ | ▼ Worse | -33% | 0% |
| case-05 | ✓→✗ | ▼ Worse | 58% | 0% |
Review the plan like a skeptical senior engineer.
You are not the user-facing coordinator for the workflow. Do not ask the user direct questions mid-review. If something needs a product or scope decision, report it as a clearly labeled recommendation for the parent workflow to aggregate after all review lanes complete.
CLAUDE.md files to understand conventionsthe source of truth for why, locked decisions, and non-goals
supporting context rather than a source of truth
Verified Repo Truths firstexisting repo patterns
Return a numbered list of recommendations. Each item must include:
Order findings by severity:
MODIFY path that does not existVerified Repo Truths that lacks exact evidenceVerified Repo Truthssilently changes a non-goal
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 7,531 | 5,360 | -29% | 1 | 1 | 0% | 437 | 1,034 | +137% | 0 | 0 | — |
case-02 | fail→fail | 22,213 | 5,337 | -76% | 1 | 1 | 0% | 4,625 | 803 | -83% | 0 | 0 | — |
case-03 | fail→fail | 8,268 | 4,956 | -40% | 1 | 1 | 0% | 1,539 | 882 | -43% | 0 | 0 | — |
case-04 | pass→fail | 19,554 | 14,290 | -27% | 1 | 1 | 0% | 4,430 | 2,984 | -33% | 0 | 0 | — |
case-05 | pass→fail | 16,219 | 20,577 | +27% | 1 | 1 | 0% | 2,622 | 4,145 | +58% | 0 | 0 | — |
case-06 | pass→pass | 6,606 | 9,121 | +38% | 1 | 1 | 0% | 1,466 | 2,403 | +64% | 0 | 0 | — |
case-07 | fail→pass | 5,670 | 6,560 | +16% | 1 | 1 | 0% | 1,034 | 1,783 | +72% | 0 | 0 | — |
case-08 | fail→fail | 6,266 | 5,159 | -18% | 1 | 1 | 0% | 351 | 778 | +122% | 0 | 0 | — |
case-09 | fail→pass | 9,778 | 10,247 | +5% | 1 | 1 | 0% | 1,719 | 1,847 | +7% | 0 | 0 | — |
case-10 | pass→pass | 12,416 | 8,435 | -32% | 1 | 1 | 0% | 2,052 | 1,761 | -14% | 0 | 0 | — |
case-11 | fail→fail | 9,017 | 4,642 | -49% | 1 | 1 | 0% | 1,565 | 759 | -52% | 0 | 0 | — |
case-12 | fail→fail | 12,594 | 7,024 | -44% | 1 | 1 | 0% | 2,122 | 748 | -65% | 0 | 0 | — |
case-13 | pass→fail | 9,078 | 6,963 | -23% | 1 | 1 | 0% | 1,482 | 989 | -33% | 0 | 0 | — |
case-14 | fail→fail | 11,683 | 8,594 | -26% | 1 | 1 | 0% | 2,055 | 995 | -52% | 0 | 0 | — |
case-15 | pass→pass | 11,105 | 9,025 | -19% | 1 | 1 | 0% | 1,838 | 2,057 | +12% | 0 | 0 | — |
case-16 | fail→pass | 7,064 | 8,591 | +22% | 1 | 1 | 0% | 1,054 | 2,133 | +102% | 0 | 0 | — |
case-17 | pass→fail | 6,343 | 5,674 | -11% | 1 | 1 | 0% | 1,025 | 960 | -6% | 0 | 0 | — |
case-18 | fail→fail | 11,189 | 5,373 | -52% | 1 | 1 | 0% | 1,993 | 957 | -52% | 0 | 0 | — |
case-19 | pass→pass | 7,021 | 10,721 | +53% | 1 | 1 | 0% | 1,120 | 2,637 | +135% | 0 | 0 | — |
case-20 | pass→pass | 11,924 | 13,136 | +10% | 1 | 1 | 0% | 2,166 | 2,465 | +14% | 0 | 0 | — |
case-21 | pass→fail | 16,671 | 5,576 | -67% | 1 | 1 | 0% | 2,758 | 807 | -71% | 0 | 0 | — |
case-22 | fail→fail | 4,653 | 5,547 | +19% | 1 | 1 | 0% | 223 | 784 | +252% | 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 10 counted toward the lift figure. The other 12 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 -9 percentage points is the difference between those two pass rates over the 10 comparable cases. 9 cases got worse with the skill loaded, and they are included in that figure.
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.