Module loop_driver 

Core agent loop driver

AgentRunner is the “missing middle” — the robust loop that sits between the apps (CLI, API) and the enact-core kernel. It implements:

• Multi-iteration tool call loop (ported from zeroclaw’s run_tool_call_loop)
• Multi-format tool call parsing (JSON, XML, Markdown)
• Auto context compaction when history grows too large
• Retry with exponential backoff for transient errors
• Checkpoint saving at configured intervals
• Stream event emission throughout
• LLM call tracing and token/cost accounting (observability)

This fulfills the mandate from 28-TS-RUST-INTEGRATION.md: “Rust handles retries/fallbacks” and 27-LONG-RUNNING-EXECUTIONS.md: “Unified long-running agentic loop”.

Structs

AgentRunner

The robust agent runner — drives the tool call loop with retries, compaction, and multi-format parsing.

Enums

LoopOutcome

Outcome of a runner loop execution.

Type Aliases

DefaultAgentRunner

Convenience: create an AgentRunner with default in-memory checkpoint store.