- Home
- Skills
- Prompt & Skill Engineering
- Tool Schema Designer: Tools Claude Calls Right
Works with the AI Agents you already use
Tool Schema Designer: Tools Claude Calls Right
This skill stops LLMs from hallucinating parameters or choosing the wrong functions by enforcing strict schema design and actionable error handling.
Free
Tool Schema Designer: Tools Claude Calls Right
Example session with this skill installed
Design a tool for a CRM agent to search for contacts. It should support filtering by name, tag, or email and handle pagination. Provide the JSON schema and test prompts.
- Read your context and instructions
- Compiled the tool schema designer
I have designed the crm_search_contacts tool using a task-oriented approach. The schema uses strict typing and provides clear examples for the LLM to follow, including specific pagination controls to prevent context overflow.
CRM Tool Definition
{
"name": "crm_search_contacts",
"description": "Searches for contacts within the CRM database based on specific filters. Use this when the user asks to find, list, or filter people. Do not use for creating or deleting contacts. Returns a maximum of 50 results per call.",
"parameters": {
"type": "object",
"properties": {
"query": {
"type": "string",
"description": "The search term for name or email. Example: 'John Doe' or 'jane@company.com'"
},
"tags": {
"type": "array",
"items": { "type": "string" },
"description": "Filter by contact tags. Example: ['prospect', 'high-value']"
},
"limit": {
"type": "integer",
"default": 20,
"description": "Number of contacts to return. Max 50."
},
"cursor": {
"type": "string",
"description": "Pagination token from a previous search result to fetch the next page."
}
},
"required": ["query"],
"additionalProperties": false
}
}
Validation Tests
| Test Prompt | Expected Tool Call | Expected Arguments | Result |
|---|---|---|---|
| "Find all prospects with the tag 'vip'" | crm_search_contacts | {"query": "", "tags": ["vip"]} | Pass |
| "Search for email support@corp.com" | crm_search_contacts | {"query": "support@corp.com"} | Pass |
| "Get the next 10 contacts" | crm_search_contacts | {"query": "", "limit": 10, "cursor": "XYZ"} | Pass |
Next steps
- Implement the tool handler to parse these JSON arguments.
- Ensure your backend returns the
cursorfield in the response metadata. - Test the "Actionable Error" path by passing an invalid limit to the LLM.
Connects securely to your tools. The creator never sees your data.
About this skill
Building reliable agentic workflows requires tools that models actually understand. This skill stops LLMs from hallucinating parameters or choosing the wrong functions by enforcing strict schema design and actionable error handling.
What it does
- Task-based architecture designs tools around specific user actions rather than just wrapping existing API endpoints.
- Strict JSON schemas defines clear parameter types with
additionalProperties: falseto prevent model confusion. - Semantic naming applies
verb_nounconventions and namespacing for clear tool selection in complex agents. - Token-efficient outputs structures returns to provide only necessary data, preventing context window bloat.
- Actionable error design replaces vague stack traces with specific instructions the LLM can use to self-correct.
How it works
- Task Inventorying creates a mapping of 3 to 15 tools based on user goals rather than backend routes.
- Schema Definition builds the JSON properties with enums, units, and clear descriptions including examples.
- Output Optimization configures response formats and pagination logic to maintain high signal-to-noise ratios.
- Validation Testing runs 10 realistic prompts to verify tool selection, argument accuracy, and token usage.
Frameworks & tools
The skill produces standard JSON Schema definitions compatible with OpenAI, Anthropic, and the Model Context Protocol (MCP). It is language-agnostic but follows best practices for Python, TypeScript, and Go SDKs.
Why this beats prompting it yourself
Generic prompts often result in tools with vague descriptions that cause LLMs to hesitate or misinterpret units. This skill follows a systematic testing loop that forces the model to prove it can call the tools correctly before you ship to production.
Use cases
- Design a suite of tools for a custom MCP server.
- Fix an agent that consistently passes incorrectly formatted dates or units.
- Refactor a bloated API wrapper into a focused set of agent tools.
- Create a standardized tool library for a multi-agent system.
Known limitations
This skill focuses on interface design and does not write the internal logic or backend implementation for the tools. It is optimized for agents using 3 to 15 tools per context.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Prompt & Skill Engineering.
Focuses on schema and interface design only. Does not write the internal execution logic or backend code. Optimized for sets of 3 to 15 tools.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
1 use
Times used to date
Free to use
No account required to browse
Trust & safety
Security scan
Security scan passed 4 days ago
- Free with an Agensi account
Permissions
No special permissions declared or detected