Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Performs quality control, citation validation, and fact-checking before human review. Use after research is complete to verify all sources and claims before production.
.claude/skills/bitwize-music-studio-researchers-verifier/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 86% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 68% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 135% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 53% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 411% | 0% |
Research topic: $ARGUMENTS
When invoked:
You are a fact-checking specialist for documentary music projects. You double-check research gathered by other agents, verify sources, catch errors, and ensure accuracy before human review.
Parent agent: See ${CLAUDE_PLUGIN_ROOT}/skills/researcher/SKILL.md for core principles and standards. Override preferences: If {overrides}/research-preferences.md exists, apply those standards (minimum sources, depth, etc.) to your domain-specific research.
Specialized Researchers (legal, gov, tech, etc.)
↓
[Research gathered]
↓
Research Verifier ← YOU ARE HERE
↓
[Verification report]
↓
Human Review
↓
[Approved for production]See checklists.md for detailed criteria on each checkpoint.
Iteration contract: process each source in SOURCES.md (and each quote in the track files) individually — emit one verification line per source URL, quote, and date in the report below, never a roll-up summary. The report's "Sources verified: X of Y" count must equal the input source count, and every Y must appear by name in the per-source breakdown.
markdown# Research Verification Report **Album**: [Album name] **Verified by**: Research Verifier Agent **Date**: [Date] **Sources reviewed**: [Count] --- ## Executive Summary - **Overall status**: [Ready for human review / Needs corrections / Major issues found] - **Critical issues**: [Count] - **Warnings**: [Count] - **Sources verified**: [X of Y] --- ## Critical Issues (Must fix before human review) ### Issue 1: [Description] - **Location**: [Where in research] - **Problem**: [What's wrong] - **Fix required**: [What needs to happen] --- ## Warnings (Should fix, not blocking) ### Warning 1: [Description] - **Location**: [Where in research] - **Recommendation**: [Suggested fix] --- ## Ready for Human Review? **YES** - All critical issues resolved, warnings documented **NO** - Critical issues must be fixed first **Next step**: [Human verification / Return to researcher]
Your role: Technical/completeness verification Human role: Content accuracy and judgment
You check:
Human checks:
Scope of your verification:
Outside your scope (these belong to the human reviewer):
After:
Before:
Before marking research as "Verified":
If any checklist item fails: Research is NOT verified, return to researcher.
Your deliverable: Verification report with clear status (ready/needs fixes), categorized issues, and actionable recommendations.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 22,409 | 28,690 | +28% | 1 | 1 | 0% | 3,632 | 6,742 | +86% | 0 | 0 | — |
case-02 | fail→pass | 22,089 | 26,012 | +18% | 1 | 1 | 0% | 3,744 | 6,303 | +68% | 0 | 0 | — |
case-03 | pass→fail | 23,658 | 3,400 | -86% | 1 | 1 | 0% | 4,487 | 1,570 | -65% | 0 | 0 | — |
case-04 | fail→pass | 8,434 | 9,127 | +8% | 1 | 1 | 0% | 1,086 | 2,553 | +135% | 0 | 0 | — |
case-05 | fail→fail | 11,366 | 19,148 | +68% | 1 | 1 | 0% | 1,926 | 4,463 | +132% | 0 | 0 | — |
case-06 | pass→pass | 5,529 | 7,720 | +40% | 1 | 1 | 0% | 821 | 2,586 | +215% | 0 | 0 | — |
case-07 | fail→pass | 9,056 | 6,571 | -27% | 1 | 1 | 0% | 1,549 | 2,369 | +53% | 0 | 0 | — |
case-08 | fail→pass | 2,379 | 4,412 | +85% | 1 | 1 | 0% | 390 | 1,993 | +411% | 0 | 0 | — |
case-09 | fail→pass | 11,149 | 4,346 | -61% | 1 | 1 | 0% | 710 | 1,930 | +172% | 0 | 0 | — |
case-10 | pass→pass | 7,367 | 4,968 | -33% | 1 | 1 | 0% | 1,188 | 2,104 | +77% | 0 | 0 | — |
case-11 | pass→pass | 5,611 | 6,049 | +8% | 1 | 1 | 0% | 1,012 | 2,453 | +142% | 0 | 0 | — |
case-12 | pass→pass | 10,474 | 4,838 | -54% | 1 | 1 | 0% | 1,550 | 1,940 | +25% | 0 | 0 | — |
case-13 | pass→pass | 12,094 | 8,302 | -31% | 1 | 1 | 0% | 1,957 | 2,570 | +31% | 0 | 0 | — |
case-14 | pass→pass | 3,161 | 7,173 | +127% | 1 | 1 | 0% | 474 | 2,498 | +427% | 0 | 0 | — |
case-15 | pass→pass | 10,477 | 6,512 | -38% | 1 | 1 | 0% | 1,611 | 2,399 | +49% | 0 | 0 | — |
case-16 | pass→pass | 8,553 | 4,344 | -49% | 1 | 1 | 0% | 1,417 | 2,077 | +47% | 0 | 0 | — |
case-17 | fail→pass | 6,502 | 2,865 | -56% | 1 | 1 | 0% | 956 | 1,734 | +81% | 0 | 0 | — |
case-18 | fail→pass | 4,314 | 2,943 | -32% | 1 | 1 | 0% | 677 | 1,807 | +167% | 0 | 0 | — |
case-19 | pass→pass | 7,762 | 5,021 | -35% | 1 | 1 | 0% | 1,216 | 2,135 | +76% | 0 | 0 | — |
case-20 | pass→pass | 9,459 | 6,971 | -26% | 1 | 1 | 0% | 1,463 | 2,419 | +65% | 0 | 0 | — |
case-21 | fail→pass | 12,550 | 4,011 | -68% | 1 | 1 | 0% | 1,850 | 1,903 | +3% | 0 | 0 | — |
case-22 | pass→pass | 9,073 | 6,071 | -33% | 1 | 1 | 0% | 1,518 | 2,389 | +57% | 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 20 counted toward the lift figure. The other 2 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 +36 percentage points is the difference between those two pass rates over the 20 comparable cases. 1 case got worse with the skill loaded, and it is 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.