Syntropy.TaskScheduler.Support (syntropy v0.2.0)

Copy Markdown

Shared helpers for the task scheduler pipeline.

Holds the small contracts used across the scheduler submodules: agent reference maps, attempt duration arithmetic, agent failure normalization, and the process-dictionary LLM usage context that ties provider calls back to the originating task and stage.

Summary

Types

agent_ref()

Functions

agent_ref(map)

Builds the minimal cross-node agent reference carried by runs and events.

current_llm_usage_context(source_review?)

Reads the LLM usage context for the calling process.

duration_ms(started_at, finished_at)

Millisecond duration between two datetimes, floored at zero.

normalize_agent_failure(reason)

Normalizes an agent execution failure into a stable error map with a machine-readable code and operator-facing message.

with_llm_usage_context(task_id, stage, fun)

Runs fun with the LLM usage process context set to task_id/stage, restoring any previous context afterwards.

Types

agent_ref()

@type agent_ref() :: Syntropy.TaskScheduler.agent_ref()

Functions

agent_ref(map)

@spec agent_ref(map()) :: agent_ref()

Builds the minimal cross-node agent reference carried by runs and events.

current_llm_usage_context(source_review?)

@spec current_llm_usage_context(boolean()) :: map()

Reads the LLM usage context for the calling process.

duration_ms(started_at, finished_at)

@spec duration_ms(DateTime.t() | nil, DateTime.t() | nil) :: non_neg_integer() | nil

Millisecond duration between two datetimes, floored at zero.

Returns nil when either endpoint is missing.

normalize_agent_failure(reason)

@spec normalize_agent_failure(term()) :: map()

Normalizes an agent execution failure into a stable error map with a machine-readable code and operator-facing message.

with_llm_usage_context(task_id, stage, fun)

@spec with_llm_usage_context(String.t(), String.t(), (-> result)) :: result
when result: var

Runs fun with the LLM usage process context set to task_id/stage, restoring any previous context afterwards.

The context is read back by current_llm_usage_context/1 when provider calls are dispatched, so token usage lands on the right task and stage.