Roadmap
This page lists planned capabilities only. Everything that works today is documented in Features. The authoritative plan lives in ROADMAP.md in the repository.
Planned means not implemented
Nothing on this page is available in the current release. Do not rely on it in tests or pipelines.
Planned capabilities
| Capability | Summary |
|---|---|
| Structured experiment reports | Every chaos run emits a report with hypothesis, injection-point provenance, observed behavior, and verdict. |
| Synthetic test scenarios | Prebuilt known-bad agent behaviors for consistent resilience validation. |
| Streaming fault injection | Fault injection at the streaming-token level for partial-stream failures. |
| Provider patching and chaos profiles | Patch an SDK client at import time so all calls go through chaos without wrapping each call manually. Declarative YAML/JSON experiment profiles. |
| Pytest plugin and assertions | First-class pytest fixtures and assertion helpers for inline resilience assertions. |
| Circuit breaker and dry-run mode | Safety rails: circuit breaker to halt chaos when error rates exceed a threshold, and a dry-run mode that reports what would fire. |
| ChaosHub | Shared experiment registry for reproducible resilience scenarios. |
Direction
Agentic Chaos is the resilience and failure-validation reference implementation for the AI Operations Workflow Specification. It aims to inject, validate, test, and prove resilience, while keeping its workflow artifacts compatible with the operational model shared by AgenticLens, the MCP layer, and the future Control Tower control plane.
Influence the roadmap
- Open an issue to request a fault type or integration.
- Read
CONTRIBUTING.mdbefore sending a pull request. - Track delivered work in the
CHANGELOG.md.