lsm / neokai / 28278429114

82%

Build Type

push

github

Committed by web-flow

Commit Message

fix(kimi): increase context reserve buffer to match Kimi's ~32k max output (#2201)

Kimi K2.7-Code / kimi-for-coding advertise ~32k max output plus mandatory reasoning tokens. The previous 33k reserve (min(maxOutputTokens, 20k) + 13k) left compaction too close to the 262k wall.

Changes:

- Make reserveBasedThreshold provider-aware: default 33k reserve, Kimi uses 45k reserve.

- Pass session providerId to reserveBasedThreshold in sdk-message-handler.

- Update context-tracker and sdk-message-handler tests for the new Kimi threshold (262144 - 45000 = 217144).

- Keep non-Kimi providers on the 33k SDK-style reserve so Anthropic/GLM/etc. are not over-compacted.

Coverage Stats

11423 of 16468 branches covered (69.36%)

Branch coverage included in aggregate %.

5 of 5 new or added lines in 2 files covered. (100.0%)

195 existing lines in 10 files now uncovered.

96529 of 115454 relevant lines covered (83.61%)

273.19 hits per line