The "context-window" topic on GitHub groups 13 open-source projects in the DeepSeek Harness (DSH) ecosystem, led by dsh-composer-history with 16 GitHub stars. dsh-composer-history — Terminal-style input history for the DeepSeek Harness web composer: edge-first arrows with exact draft/caret restore, browser-local persisted history, Ctrl+R reverse search, workspace recall - and sliding-context awareness (compaction summaries in recall/search, compaction notice with one-click /compact fill). Every project here is indexed by DSH Universe with live GitHub data — stars, activity and install status — so you can compare and install directly.
Terminal-style input history for the DeepSeek Harness web composer: edge-first arrows with exact draft/caret restore, browser-local persisted history, Ctrl+R reverse search, workspace recall - and sliding-context awareness (compaction summaries in recall/search, compaction notice with one-click /compact fill).
DeepSeek Harness MCP tool search for large catalogs: 1,000 MCP tools behind 2 MCP-facing schemas, exact-schema calls, allow/deny controls, and a local calculator.
Adds automatic context compaction, /compact, /context and model-initiated compaction to DeepSeek Harness minimal mode, fixing the problem that long tasks cannot keep running in minimal mode.
dsh-context-length is a DSH (DeepSeek Harness) settings page plugin. It adds a "Context Length" page right below "Models" in the settings panel: pick one of your configured providers, then pick a model under that provider, enter a context window value (supports 131072, 256K, 1M and similar), and click "Save" to write it into settings.yaml and take effect immediately. Unconfigured built-in providers won't appear in the list; customized models are gathered at the bottom of the page with one-click "Restore default". The whole page goes through the official settings API, the host side is always the single source of truth, safe with no side effects.
Rapid-refill circuit breaker for DeepSeek Harness automatic compaction: stops the futile compact-refill-compact loop that burns a summarization call on every step, and reports the oversized read or tool output behind it.