AI coding

Failed tool calls and error handling

• Designing Error Payloads an Agent Can Recover From
• Metering Agent Retries at the Tool Boundary
• Teaching a Tool to Return a Hint With Every Error
• A Decision Table an Agent Runs Before It Calls a Tool
• A Decision Table for Tool Errors: Fix, Retry, Escalate or Abort
• A Retry Tier Table: Errors That Get a Second Call
• Counting Wrong-Type Arguments by Tool, Field and Model
• Mapping Error Codes to Runbook Links an Agent Can Follow
• Persisting a Retry Counter Across Worker Restarts
• Repairing a Wrong-Type Argument at the Wrapper Boundary
• Retries That Cannot Succeed: A Preflight Check
• Setting Timeouts and Retries for Agent-Run Commands
• Shadow-Running a Legacy Agent: Log Schema and Cutover
• Simulating Slow and Failing Calls in a Wrapper Test
• Testing Tool Timeouts and Partial Output on Slow Calls
• When a Provider Reclassifies a Status Code Family
• When a Refusal Should Not Be Retried
• When a Wrapper Retry Belongs to the Same Task
• When a Wrapper Should Return an Error Instead of Retrying
• When an Agent Should Retry a Tool Call and When to Stop
• When Two Models Disagree on the Same Error, Which Action Wins
• Writing Error Messages a Model Can Act On