JK-LIL/dsh-compaction-current-model ↗★ 0
@dsh-external/dsh-compaction-current-model
将手动 /compact 压缩摘要请求路由至 DeepSeek Harness 前端当前选中的模型
AI 分析
核心用途是修复 DSH 0.1.0-rc.7 中手动压缩时模型选择不一致的 Bug。适合在使用手动压缩功能时,希望压缩任务严格使用前端当前所选模型的用户。
安装
$
npx -p @deepseek-ai/dsh dsh plugin --profile web add github:JK-LIL/dsh-compaction-current-model说明文档
阅读完整 README ↗@dsh-external/dsh-compaction-current-model
DSH host bundle that fixes a manual-compaction model-selection bug for
DeepSeek Harness 0.1.0-rc.7:
- Before the fix,
/compact(手动压缩)summarization could use the session's last routed model even after the frontend model picker changed. - This plugin tracks the durable manual-compaction span
(
compaction/startwithturn: null) and rewrites the summarizationllm/streamtoagentDefaultModel.currentSelection()— the model the frontend currently shows.
Automatic compaction, ordinary chat requests, and frozen LLM requests are never rewritten.
Install
dsh plugin --profile web add github:JK-LIL/dsh-compaction-current-model
or from a local tarball:
dsh plugin --profile web add ./dsh-compaction-current-model-0.1.0.tgz
then restart the web profile and run /compact after switching models.
Compatibility
- Built against DeepSeek Harness
0.1.0-rc.7(source mode or installed build). - Runtime-import-free: the plugin only calls the Cordis context and standard session/LLM events, so it has no package dependencies of its own.
- If a future DSH version changes the
compaction/startowner marker,llm/streamrequest shape, oragentDefaultModel, review this plugin before upgrading.
Uninstall
dsh plugin --profile web remove @dsh-external/dsh-compaction-current-model
Verification
From the plugin package:
npm run typecheck
npm run build
Runtime acceptance: start DSH web, switch the model picker to a model different
from the last routed request, then run /compact and confirm the recorded
compaction/summary provider/model matches the frontend selection.