Which task-list or executor phase is active for markdown generation, link validation, and (on restore) transcript replay.
Prompt / markdown: Details.to_markdown, Step.to_markdown, and List.to_markdown take a PhaseEnum to include the right sections (e.g. tool calls, result summary, references). Validation: ValidateLink and ResolveLink branch on stage so list vs refinement vs executor output get the correct rules.
Live wire ids: the skill runner adds agent-stage messages whose body is a short id (task_list_parse, refinement , exec, …). from_string maps those strings here; to_string maps back for stages the live flow emits (others yield ).
refined task block, or outstanding list (+ optional executor result summary when
exec_done) for prompts.
reference sections; see Details.to_markdown).
ResultParser.exec_extract, ResultParser.exec_post_extract, and matching
ValidateLink behavior.
it only until the first agent-stage applies.
while consuming exec_validate / validate agent-issues rows.
Details.to_markdown, List.to_markdown, ValidateLink, OLLMcoder.Skill.Runner.on_replay