dsh-context-rollover
Run the following command in DeepSeek Harness:
dsh plugin install athif23/dsh-context-rollover
Paste the following prompt into your AI chat to install this plugin:
Run dsh plugin install athif23/dsh-context-rollover in your terminal to add the dsh-context-rollover plugin to your DeepSeek Harness profile; the source is hosted at https://github.com/athif23/dsh-context-rollover — restart the host once after install to activate it.
About this plugin
Long sessions rarely fail from running out of words — they fail when the context window saturates and the system reaches for an LLM summarizer, compressing pages of nuance into a few lossy sentences. dsh-context-rollover takes a different route: the model itself decides where to close the current window and where to start a fresh one, using a deterministic checkpoint (persistent notes plus an optional handoff note) instead of a generator. Raw events stay persisted and searchable on disk, so nothing is lost to the fog of summarization.
The core tools give the model new_context (request a boundary at the next safe lifecycle point), notes (list, read, write, append, search per-session markdown), history (targeted search and read over events that have scrolled off the active surface), and get_context_remaining (an honest token-headroom reading). Beyond model-driven rollovers, the engine handles pressure-triggered compaction, provider-confirmed overflow, and the familiar /compact command — all four paths reuse the same notes-plus-verbatim-tail checkpoint so no critical context is ever dropped.
Built for developers who run multi-phase, long-horizon tasks within a single session. No network calls, no telemetry, no writes outside the notes directory, no credentials required. On the Web profile an experimental preset makes it a one-click switch; on headless profiles it activates immediately after install.
Use Cases
- Let the model pick the right moment to roll over a context window instead of relying on automatic summarization
- Keep multi-phase tasks (research to design to implementation) coherent within a single session without detail loss
- Transition safely into a fresh window via a deterministic checkpoint as the context window nears saturation
- Retrieve early-conversation details precisely after a rollover using notes and history search
Best For
- Developers running multi-phase, long-horizon reasoning tasks in a single session
- Practitioners who need verbatim, traceable context and distrust summarizer output
- DeepSeek Harness users on Web presets or headless profiles building inference workflows
Related Plugins
A unified suite combining hot runtime injection, task-aware thinking-mode routing, and a graded session protocol with red-team gates to sustain model diligence across long-horizon inference.
ModLens is a vision plugin for DeepSeek Harness that gives text-only models sight by reading images pasted directly into chat, with zero-config setup and multiple vision engines.
On-demand vision for text-only DeepSeek Harness agents: built-in free keyless vision chain and 14 vision tools, routing image turns as tool calls to vision models with pixel fidelity, no Python needed, one-command install.
Give text-only models in DeepSeek Harness eyes, enabling image Q&A, long-screenshot OCR, UI restoration, and GUI visual tasks.