Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Context-efficient tool management via search, deferred loading, and programmatic calling
.claude/skills/majiayu000-advanced-tool-use/SKILL.md| Model | Eval pass | Runs |
|---|---|---|
| gemini-3.6-flash | 96% | 24 |
| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-05 | ✗→✓ | ▲ Improved | 11% | 0% |
| case-01 | ✗→✓ | ▲ Improved | 28% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -17% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 4% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -11% | 0% |
Context-efficient tool management via search, deferred loading, and programmatic calling.
| Feature | Benefit | API Type | |---------|---------|----------| | Tool Search | 85% context reduction | tool_search_tool_regex_20251119 | | Programmatic Calling | 37% token reduction | allowed_callers: ["code_execution_20250825"] | | Tool Examples | 72%→90% accuracy | Examples in tool definition | | Deferred Loading | On-demand only | defer_loading: true |
| Category | Loading | Examples | |----------|---------|----------| | CORE | Always | health_check, get_task_status | | VERIFICATION | Deferred | verify_task, collect_evidence | | DATABASE | Deferred | query, insert | | FILE_SYSTEM | Deferred | read, write, list | | AUSTRALIAN_CONTEXT | Always | format_date_au, format_currency_aud, validate_abn |
python@dataclass class ToolConfig: defer_loading: bool = False # Load on-demand via search allowed_callers: list[str] = [] # ["code_execution_20250825"] parallel_safe: bool = True cache_results: bool = False australian_context: bool = False # Load with Australian locale
pythonfrom src.tools import register_all_tools from src.tools.search import ToolSearcher registry = register_all_tools() searcher = ToolSearcher(registry) # Search for tools (85% context savings) results = searcher.search("verify outputs", limit=3) # Get API tools with deferred loading api_tools = registry.to_api_format( include_search_tool=True, include_deferred=False, locale="en-AU" # Australian English ) # Beta header required headers = {"anthropic-beta": "advanced-tool-use-2025-11-20"}
| Metric | Before | After | |--------|--------|-------| | Upfront tokens | 55,000 | 8,000 (85%) | | Per-call tokens | 1,200 | 760 (37%) | | Parameter accuracy | 72% | 90% |
When working with Australian data:
python# Tool for formatting Australian dates @tool(australian_context=True) async def format_date_au(date: datetime) -> str: """Format date in Australian DD/MM/YYYY format.""" return date.strftime("%d/%m/%Y") # Tool for formatting Australian currency @tool(australian_context=True) async def format_currency_aud(amount: float) -> str: """Format currency in AUD with proper formatting.""" return f"${amount:,.2f}" # Tool for validating ABN (Australian Business Number) @tool(australian_context=True) async def validate_abn(abn: str) -> bool: """Validate Australian Business Number (11 digits).""" # Validation logic for ABN cleaned = abn.replace(" ", "") if len(cleaned) != 11 or not cleaned.isdigit(): return False # ABN checksum validation weights = [10, 1, 3, 5, 7, 9, 11, 13, 15, 17, 19] digits = [int(d) for d in cleaned] digits[0] -= 1 checksum = sum(d * w for d, w in zip(digits, weights)) return checksum % 89 == 0
pythonasync def execute_tools_parallel(tools: list[Tool], inputs: list[dict]) -> list[Result]: """Execute multiple tools in parallel for efficiency.""" tasks = [tool.execute(**input_data) for tool, input_data in zip(tools, inputs)] return await asyncio.gather(*tasks, return_exceptions=True)
defer_loading: trueallowed_callers for batch operationsorchestrator.get_context_stats()australian_context: true for Australian locale toolsThis skill is automatically loaded by:
.claude/agents/backend-specialist/ - For API development.claude/agents/orchestrator/ - For tool coordinationSee: tools/, verification/verification-first.skill.md
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-05 | fail→pass | 9,130 | 2,475 | -73% | 1 | 1 | 0% | 1,416 | 1,573 | +11% | 0 | 0 | — |
case-01 | fail→pass | 25,943 | 24,885 | -4% | 1 | 1 | 0% | 4,292 | 5,481 | +28% | 0 | 0 | — |
case-02 | fail→pass | 35,025 | 18,777 | -46% | 1 | 1 | 0% | 6,025 | 4,993 | -17% | 0 | 0 | — |
case-03 | fail→pass | 35,360 | 23,132 | -35% | 1 | 1 | 0% | 5,934 | 6,182 | +4% | 0 | 0 | — |
case-04 | pass→pass | 9,829 | 8,856 | -10% | 1 | 1 | 0% | 2,069 | 3,120 | +51% | 0 | 0 | — |
case-06 | fail→pass | 18,527 | 10,138 | -45% | 1 | 1 | 0% | 2,349 | 2,097 | -11% | 0 | 0 | — |
case-07 | pass→pass | 15,474 | 13,518 | -13% | 1 | 1 | 0% | 1,642 | 2,824 | +72% | 0 | 0 | — |
case-08 | fail→pass | 10,940 | 10,852 | -1% | 1 | 1 | 0% | 1,815 | 2,312 | +27% | 0 | 0 | — |
case-09 | pass→pass | 5,598 | 12,892 | +130% | 1 | 1 | 0% | 1,043 | 2,571 | +147% | 0 | 0 | — |
case-10 | pass→pass | 14,094 | 10,410 | -26% | 1 | 1 | 0% | 1,428 | 3,176 | +122% | 0 | 0 | — |
case-11 | pass→pass | 14,078 | 13,115 | -7% | 1 | 1 | 0% | 2,632 | 3,727 | +42% | 0 | 0 | — |
case-12 | fail→pass | 25,285 | 15,350 | -39% | 1 | 1 | 0% | 3,716 | 4,359 | +17% | 0 | 0 | — |
case-13 | fail→pass | 15,422 | 2,399 | -84% | 1 | 1 | 0% | 1,780 | 1,555 | -13% | 0 | 0 | — |
case-14 | fail→pass | 22,985 | 7,462 | -68% | 1 | 1 | 0% | 2,867 | 1,564 | -45% | 0 | 0 | — |
case-15 | fail→pass | 11,440 | 3,726 | -67% | 1 | 1 | 0% | 2,061 | 1,893 | -8% | 0 | 0 | — |
case-16 | pass→pass | 17,803 | 8,627 | -52% | 1 | 1 | 0% | 1,981 | 2,741 | +38% | 0 | 0 | — |
case-17 | fail→pass | 14,410 | 12,253 | -15% | 1 | 1 | 0% | 1,928 | 2,565 | +33% | 0 | 0 | — |
case-18 | fail→pass | 15,463 | 10,253 | -34% | 1 | 1 | 0% | 2,248 | 2,139 | -5% | 0 | 0 | — |
case-19 | fail→pass | 15,313 | 10,448 | -32% | 1 | 1 | 0% | 2,293 | 2,123 | -7% | 0 | 0 | — |
case-20 | pass→pass | 12,026 | 16,587 | +38% | 1 | 1 | 0% | 2,330 | 3,440 | +48% | 0 | 0 | — |
case-21 | pass→pass | 14,909 | 14,466 | -3% | 1 | 1 | 0% | 1,991 | 3,026 | +52% | 0 | 0 | — |
case-22 | pass→pass | 15,148 | 15,292 | +1% | 1 | 1 | 0% | 1,836 | 3,254 | +77% | 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. The headline lift of +59 percentage points is the difference between those two pass rates over the 22 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.