ToolCallProfile#

Create, compose, unload, and save session-local Python tools. See the guide.

class zett_agent.jit_harness.profiling.ToolCallProfile(*, call_id, tool_name, origin, status, duration_ms, input_chars=None, output_chars=None, caller_tool_name=None, error_type=None)[source]#

One finished dispatch attempt, without argument or result contents.

call_id is the join key: a model-requested call keeps the conversation’s ToolCall id, a nested call carries the id the harness assigned, and the same value appears in the JIT CUSTOM events and recent-error records.

Character counts use compact Unicode JSON for arguments/structured results, plain string length for text results, and text-block lengths for multimodal results. Image data and URLs are excluded. None means unavailable, not zero. Counts stop at the harness measurement budget, so a value equal to it means “at least this many”. Duration includes validation and nested execution, but excludes surrounding middleware, lifecycle hooks, and telemetry measurement.

Create a new model by parsing and validating input data from keyword arguments.

Raises [ValidationError][pydantic_core.ValidationError] if the input data cannot be validated to form a valid model.

self is explicitly positional-only to allow self as a field name.

Parameters:

  • call_id (str)

  • tool_name (str)

  • origin (ToolOrigin)

  • status (ToolCallStatus)

  • duration_ms (Annotated[float, Ge(ge=0)])

  • input_chars (Annotated[int | None, Ge(ge=0)])

  • output_chars (Annotated[int | None, Ge(ge=0)])

  • caller_tool_name (str | None)

  • error_type (str | None)

call_id: str#

Invocation id, shared with the conversation ToolCall and the JIT events.

tool_name: str#

Registered name that was dispatched.

origin: ToolOrigin#

Registry that served the call; see ToolOrigin.

status: ToolCallStatus#

How the attempt ended; see ToolCallStatus.

duration_ms: float#

Elapsed dispatch time in milliseconds, measured as this class documents.

input_chars: int | None#

Measured argument length, or null when the arguments are not JSON-representable.

output_chars: int | None#

Result length, or the error-message length after a failure; null when unknown or cancelled.

caller_tool_name: str | None#

Tool whose handler composed this call; null at the top level.

error_type: str | None#

Exception class name for failures and rejections; the message lives in tool_errors().

Fields#

call_id (required)

See the typed model signature for this field.

tool_name (required)

See the typed model signature for this field.

origin (required)

See the typed model signature for this field.

status (required)

See the typed model signature for this field.

duration_ms (required)

See the typed model signature for this field.

input_chars (optional)

See the typed model signature for this field.

output_chars (optional)

See the typed model signature for this field.

caller_tool_name (optional)

See the typed model signature for this field.

error_type (optional)

See the typed model signature for this field.