Skip to content

ResiliReplay v0.6.0

Latest
Latest

Choose a tag to compare

View all tags
@aliengineering-byte aliengineering-byte released this 08 Aug 02:01
· 3 commits to main since this release
v0.6.0
dd0ec88
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

ResiliReplay v0.6 adds genuine runtime reliability testing for LangGraph and the OpenAI Agents SDK—turning framework failures, recovery behavior, and broken traces into deterministic regression tests.

Highlights

  • Introduces the versioned neutral framework event contract, causal identity model, adapter SDK registry, deterministic templates, and neutral OTLP/JSONL bridge.
  • Adds genuine local runtime adapters for @langchain/langgraph@1.4.9 and @openai/agents@0.14.3 using public, provider-neutral APIs with no keys or hosted inference.
  • Adds conservative AutoGen-compatible OTLP ingestion plus documented CrewAI callback and LlamaIndex instrumentation mappings.
  • Aligns the CLI, packages, schemas, plugin, GitHub Action, documentation, and product site at 0.6.0.
  • Hardens output path containment across Windows volumes, with cross-platform CLI and security regressions.

Verification evidence

  • 23 GENUINE_RUNTIME checkpoint cases: 11 LangGraph and 12 OpenAI Agents SDK.
  • 2 FIXTURE_BACKED_PROTOCOL cases: malformed LangGraph tool-result protocol handling and AutoGen-compatible OTLP ingestion.
  • DOCUMENTED_ONLY: CrewAI event-listener and LlamaIndex instrumentation mappings.
  • Release candidate quality gate: 21 files and 143 tests passed, plus packaging, MCP Inspector, product site, field evidence, secret/privacy, and agent stress gates.
  • Release gates: 100 Studio lifecycles with zero orphan listeners, a 20,000-event trace, a four-scenario local campaign, an under-60-second demo, and a 332,086-byte CLI tarball.
  • Pull request CI passed on Node 22 and 24 across Ubuntu and Windows, with package, browser, links, and secret scans green.

Security and claim boundaries

Framework verification is local, deterministic, no-key, and bounded. It uses no paid API, hosted inference, telemetry exporter, credential access, or private monkey-patching. Evidence is sanitized and redacted; tool execution and campaign paths retain explicit allowlists, reviewed hashes, deadlines, retry bounds, path containment, and cleanup checks. Optional semantic evaluation is disabled by default and cannot override deterministic policy status.

AutoGen is protocol-fixture-backed rather than runtime-verified. CrewAI and LlamaIndex are documented mappings only. Hosted framework transports and provider billing, quota, latency, remote retry, and server-side semantics are outside this release claim. Vendor names describe interoperability and do not imply endorsement.

This is deterministic fault/recovery/regression infrastructure, not an all-in-one observability platform.

Install

npm install --global resilireplay@0.6.0
resilireplay --version

Release commit: dd0ec88a058b863fc3e8614a44f747facc76258f

Full details: CHANGELOG, framework evidence, and limitations.

Assets 2
Loading