Works with the AI Agents you already use
Agent Tool Design + Tool Definition Linter
Design the tools your LLM agent calls so it picks the right one with valid arguments first time: granularity, names, descriptions, schemas, results, errors and side-effect classes, plus a linter for tool definitions.
$9
Agent Tool Design + Tool Definition Linter
Example session with this skill installed
Our support agent keeps calling the wrong tools and dumping huge payloads into context. Review our tool definitions in tools.json and give me improved definitions in tools_v2.json plus a short review.md explaining the changes.
- Read your context and instructions
- Compiled the agent tool design
review.md: 'orders' mixes actions behind an untyped action parameter, so writes can't be gated; split into search_orders, get_order (read) and update_order_status (write-irreversible). 'get_customer' has a 13-character description and an ambiguous argument. 'list_all_tickets' returns everything: replace with search_tickets(query, status enum, limit, cursor) returning ids and snippets. tools_v2.json: 5 tools with when-to-use descriptions, required fields, additionalProperties false, enums for status, pagination; the linter reports 0 errors.
Connects securely to your tools. The creator never sees your data.
About this skill
Tools are your agent's user interface. The model only sees a name, a description, a schema and whatever text comes back, so every one of those is a prompt. Most "the model is dumb" incidents are tool-design bugs.
What your agent does with it
- Granularity: design for tasks, not endpoints; consolidate chained calls, split reads from writes, search-then-get instead of list-all, 5–20 tools per agent.
- Names and descriptions: verb_object naming with namespaces; descriptions that say what, when (and when to use a sibling), arguments with units, results, side effects, and an example.
- Schemas: required fields,
additionalProperties: false, enums, formats, bounds, flat structures, strict decoding. - Results: return what the next step needs, bounded and paginated, with explicit truncation markers.
- Errors as instructions: protocol-level error flags, what was wrong and how to fix it, retryable vs permanent vs forbidden.
- Side-effect classes (read, write-reversible, write-irreversible, external-comm) that approval policies can act on, plus idempotency keys.
- A linter (
lint_tools.py) for Anthropic, OpenAI and MCP tool formats: name style and duplicates, short descriptions, missing "when to use", untyped or undescribed properties, enum and date-format candidates, mixed read/write tools, deep nesting, list tools without pagination, oversized toolsets. JSON output for CI.
Included: SKILL.md, lint_tools.py (standard library), and an eval case with weak tool definitions to fix.
Part of the Production AI Agents series. Pair with MCP Server in Production and Agent Guardrails; start free with Agent Architecture.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for LLM Ops.
The linter checks structure and wording heuristics, not whether a tool does the right thing; pair it with selection-accuracy tests. It does not host or deploy tools (see MCP Server in Production).
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
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 1 day ago
- Software
Python 3.10+
Permissions
File Scopes
lint_tools.py reads a JSON file locally and makes no network calls.