HumanInTheLoopMiddleware[AutoModeState, Any, Any]| Name | Type | Description |
|---|---|---|
interrupt_on* | Mapping[str, bool | InterruptOnConfig] | Shared Manual interrupt map. |
worktree_root* | str | Path | Trusted repository boundary for deterministic writes. |
shell_allow_list | Sequence[str] | Default: () |
classifier_timeout_seconds | float | Default: _CLASSIFIER_TIMEOUT_SECONDS |
classifier_construction_timeout_seconds | float | Default: _CLASSIFIER_CONSTRUCTION_TIMEOUT_SECONDS |
classifier_model | str | BaseChatModel | None | Default: None |
trusted_ask_user_tool | BaseTool | None | Default: None |
trusted_compaction_tool | BaseTool | None | Default: None |
| Name | Type |
|---|---|
| interrupt_on | Mapping[str, bool | InterruptOnConfig] |
| worktree_root | str | Path |
| shell_allow_list | Sequence[str] |
| classifier_timeout_seconds | float |
| classifier_construction_timeout_seconds | float |
| classifier_model | str | BaseChatModel | None |
| trusted_ask_user_tool | BaseTool | None |
| trusted_compaction_tool | BaseTool | None |
Apply deterministic policy, classifier review, and HITL fallback.
Restrictive configured shell entries.
Timeout for one structured decision batch.
Separate timeout for lazily building a configured classifier model, so a cold provider import does not consume the inference budget.
Model the authorization classifier reviews with.
A provider:model spec is resolved lazily (and cached) on the
first review; a chat model instance is used as-is. None
inherits the main agent model, which is the default. A per-run
classifier_model on the runtime context wins over this value.
Built-in tool allowed to create consent receipts.
Built-in tool that performs conversation compaction.