Security

Tool Contracts and Schemas

• A Test Harness for Tool Selection Accuracy
• Collecting No-Tool Examples From Tickets and Call Transcripts
• Contract Tests for the Tools Your Agent Calls
• Parameterized Tools: Binding Agent Arguments to Types and Enums
• Splitting One Broad Agent Tool Into Narrow Tools
• Structured Error Schemas Across Providers
• When a Model Upgrade Forces a Tool Re-eval: A Checklist
• When a Repair Changes the Meaning: Validating Corrected Arguments
• When a Vendor Changes a Tool's Error Codes
• When to Reject a Tool Call Versus Repair It
• A Checklist for Tool Descriptions the Model Can Act On
• A Diff Tool for Tool Output Shapes Around a Vendor Change
• A Semantic Diff for Tool Arguments: Intended vs Repaired
• A Vendor Upgrade Checklist for Agents in Production
• Choosing Between a Classifier and an LLM for Tool Selection
• Cutting Token Spend by Trimming Tool Schemas
• Denial as Input: What the Agent Sees After a Refused Call
• Generating Tool Schemas from Database Columns
• Keeping Control Tests Current When Your Agent's Tools Change
• Keeping Eval Sets Current as Tools Are Added and Retired
• Mapping Your Agent's Tools to the Data They Can Reach
• Near-Miss Argument Pairs: A Test Corpus for Repairs
• No-Tool Classifier: When to Check Abstention Before the Model
• Pinning Error Codes in Your Client Library, Not Your Config
• Reading Repair Rates as a Signal, Not a Score
• Repair Rules That Read the Unit, Not the Number
• Returning Errors the Model Can Recover From
• Reviewing a Tool Contract When a Vendor Changes the API
• Sandboxing Agent Tools: Filesystem, Network and Shell Limits
• Tool Contracts Across Model Providers: One Schema, Many Formats
• Tool Result Envelopes That Survive a Provider Swap
• Tool Schemas That Only Allow Meaning-Preserving Repairs
• Typed Tool Definitions: MCP, OpenAI and Anthropic
• Validating Tool Outputs Against a Schema Before the Agent
• Versioning a Tool Contract So Repair Code Knows What It Targets
• What the Model Attends to in a Tool Error Message
• What the Model Sees When a Tool Returns an Error
• When the Tool Description Causes the Failure
• Where the Unit Lives: Schemas That Carry Currency and Measure
• Wording a Refusal the Model and the User Both Understand
• Writing a Tool Description That Makes the No-Tool Case Obvious
• Writing the Contract for Every Tool an Agent Can Call
• Writing the Tool Description the Model Reads
• Writing Tool Contracts That Make Repair Obvious