agentQuake: test what your AI SDK agent does when its tools fail

Hi all! I built agentQuake, an open source library that makes an agent's tools fail on purpose inside your Vitest or Jest tests, and reports how the agent handled it.

The case that got me: a tool returns 200 with an empty body, nothing throws, and the agent happily replies "Forecast for undefined". agentQuake injects that kind of silent fault (plus 503s, timeouts and latency) and flags:

falseSuccess: a tool failed, nothing recovered, and the answer still reads like a success
leakedGarbage: undefined/null/NaN from a corrupted tool ended up in the reply

The repo has an AI SDK example (ToolLoopAgent) with a naive and a careful agent. It runs without an API key: npm ci && npm run example

Repo: https://github.com/saeedallam/agentquake Write-up: https://dev.to/saeed_allam_5f71578da546b/my-ai-agent-said-forecast-for-undefined-and-no-test-caught-it-so-i-built-a-chaos-testing-tool-2adk

It's early (0.4), so I'd love feedback from people running AI SDK agents in production: what tool failures have bitten you?

Hi all! I built agentQuake, an open source library that makes an agent's tools fail on purpose inside your Vitest or Jest tests, and reports how the agent handled it.

falseSuccess: a tool failed, nothing recovered, and the answer still reads like a success
leakedGarbage: undefined/null/NaN from a corrupted tool ended up in the reply

The repo has an AI SDK example (ToolLoopAgent) with a naive and a careful agent. It runs without an API key: npm ci && npm run example

Repo: https://github.com/saeedallam/agentquake Write-up: https://dev.to/saeed_allam_5f71578da546b/my-ai-agent-said-forecast-for-undefined-and-no-test-caught-it-so-i-built-a-chaos-testing-tool-2adk

It's early (0.4), so I'd love feedback from people running AI SDK agents in production: what tool failures have bitten you?