reaatech/agent-chaos
These packages provide a fault injection toolkit for testing the resilience of LLM agent systems against real-world failures like latency, rate limits, and malformed tool outputs. You would adopt these to validate that your circuit breakers, fallbacks, and error-handling logic function correctly when external dependencies fail. The system uses a transparent middleware architecture that intercepts tool calls, allowing you to inject faults declaratively via YAML or JSON without modifying your agent's core implementation.
Packages
6 packages
@reaatech/agent-chaos-adapters
Injects fault-testing logic into agent frameworks like LangChain, LlamaIndex, and Vercel AI SDK by wrapping tool calls through a transparent interceptor pattern. It requires `@reaatech/agent-chaos-core` to execute the underlying fault injection engine.
- status
- awaiting publish
@reaatech/agent-chaos-cli
Provides a CLI and programmatic API for managing, validating, and executing chaos engineering scenarios for AI agents. It supports project initialization, template generation, and hot-reloading execution of fault-injection scenarios defined in JSON or YAML.
- status
- published
- published
- 8 days ago
@reaatech/agent-chaos-core
Injects simulated failures into agent tool calls using a middleware-based engine that supports scenario-driven configurations and deterministic, seeded randomness. It provides a `createChaosEngine` factory function that returns an instance with an `intercept` method for wrapping agent execution logic.
- status
- published
- published
- 8 days ago
@reaatech/agent-chaos-e2e
Provides an end-to-end test suite for verifying the integration of the agent-chaos pipeline, including scenario loading, fault injection, and CLI execution. It is designed to run against mock agent implementations to validate cross-package functionality.
- status
- awaiting publish
@reaatech/agent-chaos-observability
Collects metrics, OpenTelemetry traces, and structured logs for fault injection experiments performed by the @reaatech/agent-chaos-core engine. It provides a suite of classes for exporting observability data to monitoring providers and generating experiment reports in JSON, HTML, or JUnit XML formats.
- status
- awaiting publish
@reaatech/agent-chaos-scenarios
Provides a `ScenarioLoader` class and `SchemaValidator` utility to parse, validate, and hot-reload YAML or JSON chaos injection scenarios. It supports scenario composition via inheritance and acts as the configuration management layer for the `agent-chaos` toolkit.
- status
- published
- published
- 8 days ago
Comments
Sign in with GitHub to comment and vote.
Loading comments…