LangChain Releases Automated Eval Skill with Multi-Turn Simulation
LangChain · x · 2026-07-31
LangChain has released a new version of its Eval Engineering Skill for coding agents, designed to help automate the creation of evaluations using repository context and agent traces.
Key updates in this release include:
- Multi-turn Simulation: Now supports simulating users in multi-turn evaluations using custom instructions and examples.
- Environment Design: Offers expanded guidance on backend/database simulation and using traces to ensure the test environment mirrors production.
- Iterative Refinement: Added instructions for iteratively refining the evaluation harness, environment, and verifier.
The skill can be integrated into coding agents that support the Agent Skills specification (such as Claude Code and Cursor) via npx skills add.
More from coding & agent
- Woken Up at Night: Codex Autonomously Launches Emulator to Test Game — billyjhowell · 2026-07-31
- LangSmith LLM Gateway Enters Public Beta for Unified Agent Governance — LangChain · 2026-07-31
- Ditching Enumerated APIs: Modem Shares MCP Server Engineering Practices — zeeg · 2026-07-31
- OpenWiki v0.2.4 Adds Multilingual Support for Agent Wikis — hwchase17 · 2026-07-31
- Via Higgsfield MCP, Users Can Generate Videos Directly in Claude — socialwithaayan · 2026-07-31
- OpenAI Upgrades Auto-review to GPT-5.6 Luna, Cutting Costs 10x — OpenAI · 2026-07-31