They break it because nobody checks in real time, against the actual statute.
https://skillsmd.si verifies every rule verbatim against US law, then watches the sources for changes.
I'm working on making it the system that will help you handle swarms of agents, modifying the skills.md file when needed and cancelling underperforming agents.
Here is what found someone at Anthropic published today -->
Verifying rules verbatim against the source text is a solid angle, since agents tend to paraphrase statutes confidently and get them wrong. How do you handle a statute that changes mid-run for an agent already working?
Hi Vineet, I'm new here.
Just getting connected again to the platform.
Thanks for the question.
I agree.
To refine the behavior of an agent that is already working we can update programatically via the supervisor agent the skills.md file for the agent. So the supervisor...
Solid list, and the duplicate-action check is the one most agents miss. When I build MCP servers in Go, the thing that helps most is an idempotency key on every write tool. If the server sees the same key twice, it returns the first result, so a retry can't double-book or...
An AI agent fails halfway through a job. What should it remember when it tries again?
That question led me to build AgentTrace, a Python and FalkorDB project for keeping track of agent runs. It records task progress and decisions in a graph so you can inspect what happened and see where a run stopped. That’s especially useful when several agents share a workflow.
When you debug an agent run, what do you look for first: the failed step, the input that caused it, or the decision that sent it down that path?