Runner, a local LLM engine, handles tool calls that are truncated by token limits by completing them as valid JSON and returning a parseable result, allowing agent loops to continue instead of failing or retrying from scratch like other engines.
Researchers discovered that malicious LLM routers can intercept and modify tool calls before agent execution, with one attack draining a client's $500k wallet. OrcaRouter proposes an Agent Firewall defense that validates tool calls at the execution boundary, offering allow, audit, deny, sanitize, approval hold, and cost-cap capabilities.