Tool Search Skill
Use this skill when an agent has many tools available and does not know the exact tool name or input schema upfront. The skill keeps tool discovery cleanly separated from tool execution — which prevents the most common failures such as “tool not found” or invalid parameters.
---name: tool-search-skilldescription: Always use this skill whenever you need to find, select, and execute a tool for a specific task but do not yet know its exact name or input schema.---
# Tool Search Skill
Finds the right tool via tool search, selects it based on name and schema, and then executes it with the correct parameters.
## Step 1: Discover Available Tools
Call the `search_tools` (or `tool_search`) utility first to find out which tools exist.
- **The query rule:** The `query` parameter MUST be a generic, functional keyword describing *what* you want to achieve (e.g. `"search"`, `"files"`, `"excel"`, `"database"`, `"sql"`, `"image"`).- **CRITICAL – Do NOT use the user's content search query.** Do not search for `"machine learning"` or `"maschinelles Lernen"`.- **CRITICAL – Do NOT pass tool-specific arguments** (such as `collection`, `topK`, `searchMode`, `mcp_server`) to the tool discovery utility. Only use the arguments defined in the discovery tool's schema.
Example call:
```json{ "query": "search"}```
## Step 2: Analyse and Select the Correct Tool
Review the list of tools returned by your discovery search.
1. **Identify the exact name:** Look at the `name` field of the matched tool. You must use the full, exact name (e.g. `search_content_mcp_ai-search`, NEVER truncate it to `search_content`).2. **Locate the schema:** Read the tool's description and parameter definitions to understand which parameters it expects.
## Step 3: Execute the Target Tool
Invoke the selected tool.
- Only in this step do you apply the user's actual search query, filters, and operational parameters (e.g. `collection: "ai_rag_ml"`, `searchMode`, `query`).- Match the input schema of the target tool precisely. Do not carry over any parameters from the discovery step.
## Core Rules
| Rule | Meaning || --- | --- || **Strict parameter separation** | Never mix discovery parameters (the discovery query) with operational parameters (`collection`, `topK`, …). || **Query distinction** | A functional tool keyword (e.g. `"search"`) is not the user's content query (e.g. `"machine learning"`). || **Exact name matching** | Copy the full tool name character by character to prevent "tool not found" errors. |