refactor(claude-kiro): improve token counting accuracy using API contextUsagePercentage

Browse source

Replace client-side token estimation with server-provided contextUsagePercentage
for accurate input token calculation. This eliminates discrepancies between
estimated and actual token counts by using the API's native context window
measurement.

Key changes:
- Add contextUsage event parsing in AWS Event Stream buffer
- Calculate input tokens from contextUsagePercentage in both streaming and
  non-streaming responses
- Deprecate estimateInputTokens() method in favor of API-provided metrics
- Import CLAUDE_DEFAULT_MAX_TOKENS for context window calculations
- Add fallback error handling when contextUsagePercentage is not received
- Fix code formatting and whitespace consistency throughout the file

...

This commit is contained in:

Zhafron Kautsar 2026-01-07 23:39:07 -05:00

parent ea2ae2169c

commit 89bc0ee6f8

1 changed files with 359 additions and 281 deletions

Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL

Show stats Download patch file Download diff file Expand all files Collapse all files

640

src/claude/claude-kiro.js

View file

File diff suppressed because it is too large Load diff