ๆญฃๅจๅ ่ฝฝ่ง้ข...
่ง้ขๅ ่ฝฝๅคฑ่ดฅ
๐ Announcing LangSmith Skills + CLI ๐ Agent improvements are increasingly driven by coding agents themselves. We're releasing LangSmith Skills alongside the LangSmith CLI to make coding agents experts at the agent engineering lifecycle. LangSmith Skills enable agents to debug traces, create datasets, and run experiments - and thanks... show more
49,662 ๆฌก่ง็ โข 7 ไธชๆๅ โขvia X (Twitter)
12 ๆก่ฏ่ฎบ
@hwchase17 piping LangSmith traces directly into your coding agent so it debugs your other agents โ agents improving agents is a literal workflow now. CLI-first was the right call
This is the future of agent engineering. Key insight: when agents can debug their own traces via CLI, you get compounding improvements. My checklist for agentic workflows: 1) instrument every step with trace IDs, 2) build feedback loops where agents learn from failed runs, 3) keep agents in terminal where they're most productive. The 50% VRAM reduction from Unsloth + LangSmith = serious edge deployment potential.
Nice
the toolchain is eating itself. agents improving agents. the question is whether this feedback loop converges on better software or just more software. best stacks will be the ones that know when to stop generating.
check out to get sqry, ships as an MCP server (33 tools). Claude Code, Codex, Gemini can query the actual call graph instead of guessing. make @LangChain smarter.
Gonna try it Looks good
Thank you! I was making do with the locally hosted MCP server and it was quite finicky.
Agents improving agents via traces + CLI feels like the toolchain finally eating itself in the best way. One thing I'm wondering: how are you all thinking about stopping conditions or 'convergence' in these self-improvement loops so it doesn't just keep generating more code forever?
We're getting closer to agents running the full engineering loop: build โ debug โ eval โ improve.
Awesome, would be cool to get some example usages. How do i get claude code running to fix a specific langgraph call given the langsmith run/trace id?
Agents improving their own traces via CLI is a huge step. Self-improving workflows > one-shot execution. We're building on this vision at @getvibecodes โ AI agents as persistent team members across the full dev lifecycle. LangSmith Skills makes the ecosystem stronger for everyone.
The AI slop replies on langchain X posts are unbearable
