Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Create and manage sandboxed WASM agents for isolated code execution
.claude/skills/majiayu000-wasm-agent/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-07 | ✗→✓ | ▲ Improved | -68% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -66% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -69% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -68% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -53% | 0% |
Create sandboxed agents that run in WebAssembly for safe, isolated execution.
When you need to run untrusted code, experiment with agent configurations, or create portable agents that run anywhere WASM is supported.
mcp__claude-flow__wasm_agent_create with agent configurationmcp__claude-flow__wasm_agent_prompt to interact with the agentmcp__claude-flow__wasm_agent_tool to give the agent access to specific toolsmcp__claude-flow__wasm_agent_files to read/write files in the sandboxmcp__claude-flow__wasm_agent_export to package the agent for sharingmcp__claude-flow__wasm_agent_list to see all running WASM agentsmcp__claude-flow__wasm_agent_terminate to stop an agent| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-05 | pass→pass | 15,651 | 14,747 | -6% | 1 | 1 | 0% | 1,859 | 1,827 | -2% | 0 | 0 | — |
case-06 | pass→pass | 8,314 | 5,285 | -36% | 1 | 1 | 0% | 1,482 | 1,219 | -18% | 0 | 0 | — |
case-01 | fail→fail | 15,667 | 19,489 | +24% | 1 | 1 | 0% | 1,772 | 901 | -49% | 0 | 0 | — |
case-02 | fail→fail | 8,689 | 11,674 | +34% | 1 | 1 | 0% | 1,248 | 642 | -49% | 0 | 0 | — |
case-03 | fail→fail | 19,476 | 9,282 | -52% | 1 | 1 | 0% | 2,169 | 990 | -54% | 0 | 0 | — |
case-04 | pass→pass | 10,049 | 13,295 | +32% | 1 | 1 | 0% | 1,686 | 1,870 | +11% | 0 | 0 | — |
case-07 | fail→pass | 17,891 | 3,365 | -81% | 1 | 1 | 0% | 2,255 | 715 | -68% | 0 | 0 | — |
case-08 | fail→pass | 18,911 | 5,179 | -73% | 1 | 1 | 0% | 2,079 | 701 | -66% | 0 | 0 | — |
case-09 | fail→pass | 15,603 | 2,372 | -85% | 1 | 1 | 0% | 2,510 | 768 | -69% | 0 | 0 | — |
case-10 | fail→pass | 13,954 | 7,767 | -44% | 1 | 1 | 0% | 2,006 | 639 | -68% | 0 | 0 | — |
case-11 | fail→pass | 12,277 | 1,825 | -85% | 1 | 1 | 0% | 1,182 | 559 | -53% | 0 | 0 | — |
case-12 | fail→pass | 13,831 | 8,106 | -41% | 1 | 1 | 0% | 1,621 | 885 | -45% | 0 | 0 | — |
case-13 | fail→pass | 19,429 | 3,776 | -81% | 1 | 1 | 0% | 3,118 | 806 | -74% | 0 | 0 | — |
case-14 | fail→pass | 17,009 | 13,219 | -22% | 1 | 1 | 0% | 2,928 | 2,094 | -28% | 0 | 0 | — |
case-15 | fail→pass | 7,281 | 6,739 | -7% | 1 | 1 | 0% | 1,310 | 644 | -51% | 0 | 0 | — |
case-16 | fail→pass | 16,879 | 7,804 | -54% | 1 | 1 | 0% | 2,716 | 809 | -70% | 0 | 0 | — |
case-17 | fail→pass | 7,740 | 8,317 | +7% | 1 | 1 | 0% | 1,303 | 709 | -46% | 0 | 0 | — |
case-18 | fail→fail | 18,295 | 8,133 | -56% | 1 | 1 | 0% | 2,321 | 836 | -64% | 0 | 0 | — |
case-19 | fail→pass | 15,525 | 2,527 | -84% | 1 | 1 | 0% | 1,793 | 774 | -57% | 0 | 0 | — |
case-20 | fail→fail | 12,437 | 2,759 | -78% | 1 | 1 | 0% | 1,242 | 790 | -36% | 0 | 0 | — |
case-21 | fail→pass | 6,936 | 3,182 | -54% | 1 | 1 | 0% | 1,249 | 934 | -25% | 0 | 0 | — |
case-22 | fail→pass | 10,208 | 3,054 | -70% | 1 | 1 | 0% | 1,778 | 930 | -48% | 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 +64 percentage points is the difference between those two pass rates over the 20 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.