CATEGORY
Verified manifests and exact versions in this category.
@local/model-thinking-levelsExpose per-model thinking levels (reasoningEfforts) inside the Models settings card for a custom model API route.
dsh-adaptive-max-tokensAdaptive output-token budgeting for DeepSeek Harness: shrink maxTokens as context fills, restore it after compaction, and retry recognized output-budget overflows once.