apunkt/rowboat
1
0
Fork 0
mirror of https://github.com/rowboatlabs/rowboat.git synced 2026-07-12 21:02:17 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

feat(core): lower default tool-result elision threshold to 2500 chars

Browse source 
Real session data shows ~5k-char skill bodies are the most common
oversized historic result; the 10k default replayed them on every model
call for the life of the session. With the head preview in place the
model retains enough scent to re-load on demand, so the aggressive
default is the right trade. Still tunable via config/context.json.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
Ramnique Singh 2026-07-07 20:23:18 +05:30
parent 04348dcc41
commit 331a5eda4e
2 changed files with 6 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

2
apps/x/packages/core/docs/turn-runtime-design.md
View file

@ -405,7 +405,7 @@ byte-stable across calls (provider prefix caches keep working), and the
current turn's own messages never pass through the resolver, so in-flight current turn's own messages never pass through the resolver, so in-flight
tool results and just-captured frames are always sent verbatim. Policy lives tool results and just-captured frames are always sent verbatim. Policy lives
in `config/context.json` (`elideHistoricToolResults`, default true; in `config/context.json` (`elideHistoricToolResults`, default true;
`elideHistoricToolResultsThresholdChars`, default 10000; `elideHistoricToolResultsThresholdChars`, default 2500;
`elideHistoricImages`, default true; `elideHistoricMiddlePaneContent`, `elideHistoricImages`, default true; `elideHistoricMiddlePaneContent`,
default true). The inspect CLI composes through the same decorated resolver. default true). The inspect CLI composes through the same decorated resolver.

6
apps/x/packages/core/src/turns/context-elision.ts
View file

@ -43,9 +43,13 @@ export interface ElisionPolicy {
middlePaneContent: boolean; middlePaneContent: boolean;
} }
// Threshold rationale: ~2,500 chars ≈ 600 tokens. Observed sessions show
// skill bodies at ~5k chars are the most common oversized result, so 10k
// would replay them forever; the head preview plus re-run hint makes the
// lower cut safe. Tunable via config/context.json.
export const DEFAULT_ELISION_POLICY: ElisionPolicy = { export const DEFAULT_ELISION_POLICY: ElisionPolicy = {
toolResults: true, toolResults: true,
toolResultThresholdChars: 10_000, toolResultThresholdChars: 2_500,
images: true, images: true,
middlePaneContent: true, middlePaneContent: true,
}; };