llm-agent-tool-calling-optimization

What problem does it solve?

LLM agents often perform iterative, sequential searches when calling server-side tools, leading to longer response times. This Skill enforces parallel execution and explicit tool call budgets to keep latency predictable.

Core Features & Use Cases

• Parallel execution: Constrain tool calls so multiple tools run in a single batch rather than in series.
• Budgeted tool calls: Define exact limits on total tool calls across all tools.
• Deterministic planning: Require upfront planning of queries and avoid follow-up searches based on initial results.
• Use Case: In a product-search agent, consolidate multiple web searches into one parallel query to return results faster and within budget.

Quick Start

• Add to the system prompt: "Search Constraints (CRITICAL) - Maximum N total searches: up to X tool_a + up to Y tool_b; Plan ALL search queries upfront, then execute them in ONE parallel batch; Do NOT add follow-up searches based on initial results; Use OR operators to combine intents."
• Example consolidation: Turn separate queries like "TOKEN launch", "TOKEN migration", "TOKEN listing", "TOKEN partnership" into "TOKEN (launch OR migration OR listing OR partnership)".
• Time constraints: If the SDK supports a time range (from_date, to_date), use it directly; otherwise add soft constraints in the prompt such as "Focus on events from {search_from_date} onwards" and "Skip results before {search_from_date}".