Expand description
No-LLM proactive pruning of old tool results by count.
Runs before crate::compact::Compactor (LLM-driven compaction) so that
after pruning tool results by count, the transcript may drop below the
char-based compaction threshold and the expensive LLM summary is skipped
entirely. See Microcompactor::prune for details.
Structs§
- Microcompactor
- Configuration for count-based proactive pruning of old tool results.
Constants§
- MICROCOMPACT_
PLACEHOLDER - Placeholder reused from
run_core::TRIM_PLACEHOLDERsemantics. Re-exported here so the microcompactor and the char-budget trim use the same marker text. - MIN_
PRUNE_ LENGTH - Minimum tool-result size worth pruning; shorter results are kept verbatim
(matches
run_core::MIN_TRIM_LENGTH).
Functions§
- build_
microcompactor_ from_ env - Build a
Microcompactorfrom environment-variable strings.