Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Test skills via TDD in fresh subagents. Use when validating behavior or preventing bias.
.claude/skills/athola-subagent-testing/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-06 | ✗→✓ | ▲ Improved | -52% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -20% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -16% | 0% |
| case-12 | ✗→✓ | ▲ Improved | -35% | 0% |
| case-14 | ✗→✓ | ▲ Improved | -22% | 0% |
Test skills with fresh subagent instances to prevent priming bias and validate effectiveness.
abstract:skill-authoring)abstract:skills-eval)Fresh instances prevent priming: Each test uses a new Claude conversation to verify the skill's impact is measured, not conversation history effects.
Running tests in the same conversation creates bias:
Three-phase TDD-style approach:
Test without skill to establish baseline behavior.
Test with skill loaded to measure improvements.
Test skill's anti-rationalization guardrails.
bash# 1. Create baseline tests (without skill) # Use 5 diverse scenarios # Document full responses # 2. Create with-skill tests (fresh instances) # Load skill explicitly # Use identical prompts # Compare to baseline # 3. Create rationalization tests # Test anti-rationalization patterns # Verify guardrails work
For complete testing patterns, examples, and templates:
without the skill active, with full response text recorded.
of the baseline conversation) and shows >= 50% improvement on skill-related metrics.
attempts tested across at least 3 pressure scenarios.
outcomes, confirming the effect is not conversation-history priming.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-02 | fail→fail | 28,236 | 21,764 | -23% | 1 | 1 | 0% | 3,559 | 4,429 | +24% | 0 | 0 | — |
case-01 | fail→fail | 19,134 | 25,713 | +34% | 1 | 1 | 0% | 3,318 | 4,536 | +37% | 0 | 0 | — |
case-03 | fail→fail | 18,595 | 3,900 | -79% | 1 | 1 | 0% | 2,958 | 1,072 | -64% | 0 | 0 | — |
case-04 | fail→fail | 16,861 | 9,071 | -46% | 1 | 1 | 0% | 2,488 | 2,184 | -12% | 0 | 0 | — |
case-05 | fail→fail | 17,264 | 10,563 | -39% | 1 | 1 | 0% | 2,644 | 2,281 | -14% | 0 | 0 | — |
case-06 | fail→pass | 16,183 | 4,404 | -73% | 1 | 1 | 0% | 2,929 | 1,420 | -52% | 0 | 0 | — |
case-07 | pass→pass | 17,996 | 2,717 | -85% | 1 | 1 | 0% | 2,727 | 1,137 | -58% | 0 | 0 | — |
case-08 | pass→pass | 12,392 | 4,874 | -61% | 1 | 1 | 0% | 1,736 | 1,465 | -16% | 0 | 0 | — |
case-09 | fail→pass | 10,326 | 3,131 | -70% | 1 | 1 | 0% | 1,560 | 1,248 | -20% | 0 | 0 | — |
case-10 | pass→pass | 18,072 | 5,585 | -69% | 1 | 1 | 0% | 2,635 | 1,581 | -40% | 0 | 0 | — |
case-11 | fail→pass | 10,621 | 4,037 | -62% | 1 | 1 | 0% | 1,593 | 1,345 | -16% | 0 | 0 | — |
case-12 | fail→pass | 12,133 | 2,675 | -78% | 1 | 1 | 0% | 1,871 | 1,213 | -35% | 0 | 0 | — |
case-13 | pass→pass | 15,454 | 9,302 | -40% | 1 | 1 | 0% | 2,372 | 2,129 | -10% | 0 | 0 | — |
case-14 | fail→pass | 13,619 | 5,340 | -61% | 1 | 1 | 0% | 1,928 | 1,506 | -22% | 0 | 0 | — |
case-15 | pass→pass | 13,853 | 6,379 | -54% | 1 | 1 | 0% | 2,015 | 1,687 | -16% | 0 | 0 | — |
case-16 | fail→pass | 14,819 | 2,870 | -81% | 1 | 1 | 0% | 2,252 | 1,234 | -45% | 0 | 0 | — |
case-17 | fail→pass | 8,036 | 3,236 | -60% | 1 | 1 | 0% | 1,162 | 1,267 | +9% | 0 | 0 | — |
case-18 | fail→pass | 10,190 | 6,962 | -32% | 1 | 1 | 0% | 1,599 | 1,753 | +10% | 0 | 0 | — |
case-19 | fail→pass | 11,659 | 4,294 | -63% | 1 | 1 | 0% | 1,716 | 1,381 | -20% | 0 | 0 | — |
case-20 | pass→pass | 8,315 | 3,261 | -61% | 1 | 1 | 0% | 1,212 | 1,291 | +7% | 0 | 0 | — |
case-21 | fail→pass | 12,235 | 9,974 | -18% | 1 | 1 | 0% | 1,977 | 2,404 | +22% | 0 | 0 | — |
case-22 | pass→pass | 11,070 | 2,439 | -78% | 1 | 1 | 0% | 1,596 | 1,114 | -30% | 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 +45 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.