Coder/deepseek-harness
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge master into compact-tool-result-prune

Browse Source
This commit is contained in:
Tianyi Cui
2026-07-20 16:41:59 +08:00
parent 1cd999b3fe 34c0800d3c
commit 6b44d01bc8
807 changed files with 15501 additions and 3177 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
packages/compact/compact-tool-result-prune/README.md
View File

@@ -39,9 +39,17 @@ export function apply(ctx: Context): void {
### Pruned tool result
**What the model sees**: Once a compaction trigger qualifies, future requests see the retained head, `\n\n[... tool result middle pruned ...]\n\n`, and retained tail in place of the removed text. Rich blocks keep their order. The model does not see a second copy of the original.
#### What the model sees
**Token effect**: Each rewritten tool result has at most `thresholdChars` text code points. Pruning itself makes no model call; compact-basic skips summarization when the remeasured request falls below pressure, otherwise the summarizer reads the pruned surface.
Once a compaction trigger qualifies, future requests see the retained head, `\n\n[... tool result middle pruned ...]\n\n`, and retained tail in place of the removed text. Rich blocks keep their order. The model does not see a second copy of the original.
#### Token effect
Each rewritten tool result has at most `thresholdChars` text code points. Pruning itself makes no model call; compact-basic skips summarization when the remeasured request falls below pressure, otherwise the summarizer reads the pruned surface.
#### KV Cache effect
Replacing an earlier result invalidates the reusable request prefix from that result onward; subsequent requests reuse the new pruned prefix until another surface replacement occurs.
## Known Limitations and Deferred Work