The executing agent (per-agent compaction knobs + summary prompt override)
OptionalAgentThe executing agent run's ID — stamped onto the summary AIPromptRun (AgentRunID) so per-run cost rollups include the summary sub-call this feature generates.
The agent's type (type-level defaults) — null only for typeless programmatic runs
The resolved budget (see ConversationCompactionManager.ResolveEffectiveBudget)
The conversation whose window may need compacting
Token estimator for context messages — supplied by the caller so trigger math uses the SAME heuristic BaseAgent uses for in-turn context management (no drift).
OptionalExcludeDetail rows to omit from the assembled window — e.g. the in-flight agent-response placeholder row, whose Message the resolver may still be writing concurrently with a post-turn pass. Without this, the boundary could land on that row and the summary write's Load→Save (generated spUpdate writes all columns) could clobber it.
OptionalOnFires with the summary AIPromptRun ID as soon as the run record exists (step TargetLogID wiring)
OptionalProviderProvider for the boundary-row entity write (falls back to the global provider)
OptionalVerboseVerbose diagnostics (LogStatusEx verboseOnly)
Inputs to ConversationCompactionManager.CompactIfNeeded.