yin1245 88523e099c fix: use per-model context window size for accurate input_tokens estimation ... Different Claude models have different context window sizes: - Opus 4.5/4.6: 1M tokens - Sonnet 4.5/4.6: 200K tokens - Haiku 4.5: 200K tokens The current implementation uses a single hardcoded TOTAL_CONTEXT_TOKENS constant for all models, which causes input_tokens to be overestimated for small-context models or underestimated for large-context models. This change adds a MODEL_CONTEXT_TOKENS mapping and a helper function getContextTokensForModel() that selects the correct context window size based on the model being used, falling back to KIRO_CONSTANTS.TOTAL_CONTEXT_TOKENS for unknown models.		2026-03-29 12:31:46 +08:00
..
claude-core.js	feat: 增强TLS sidecar支持并更新模型列表	2026-03-16 18:26:16 +08:00
claude-kiro.js	fix: use per-model context window size for accurate input_tokens estimation	2026-03-29 12:31:46 +08:00
claude-strategy.js	feat(logging): 添加日志系统配置和下载功能	2026-01-25 17:24:39 +08:00