dsh-force-compact
Aggressive context compaction and thinking-control plugin for DeepSeek Harness local-first agents.
Install
$ dsh plugin --profile web add github:falling-ts/dsh-force-compactPlugin Overview & Capabilities
AI-assisted organization based on the public repository snapshot. The content must be grounded in source evidence and does not replace compatibility or security verification.
Key Capabilities
- Monitors projected context token count on agent/pre-step hooks to trigger compaction before model requests exceed thresholds
- Preserves the most recent tokens verbatim while summarizing older message spans into structured checkpoint nodes
- Provides automatic fallback to an internal builtin compaction transaction when the official compaction service is unreachable
- Selectively turns off reasoning/thinking during compaction summarization calls while preserving thinking configuration for main tasks
- Offers turn-end idle compaction, manual /force-compact slash commands, and durability checkpoint integration (session/flush)
- Includes a web client UI with real-time status indicators in running labels, composer context ring markers, and per-session threshold overrides
Useful For
- Running deep conversations on low-context local LLM backends (e.g., llama.cpp) without hitting context length limits
- Minimizing local inference latency and token overhead by keeping ongoing prompt history lean
- Maintaining long-running agent workflows without external API egress or token costs
Who It Fits
- Developers running local-first LLMs via llama.cpp or OpenAI-compatible backends within DeepSeek Harness
- Users needing long-session memory retention within strict token limits
Documented Limitations
- Minimum threshold floor is capped at 32,000 tokens for auto-compaction and 8,000 tokens for retained latest tokens
- The /force-compact slash command loads lazily and is only registered process-wide after the session initiates its first model request
DSH Compatibility
Version-specific runtime evidence collected by DSH Plugin. A missing result means we have not tested that combination yet.
Security Signals
Objective signals discovered from package metadata and source inspection. These are not a guarantee that a plugin is safe.
package.json declares 0 runtime, 0 development, 14 peer, and 0 optional dependencies.
package.json declares DSH bundle metadata.
GitHub reports the repository license as MIT.
A root package.json was captured and can be inspected by the registry.
Public GitHub source metadata is available for this registry snapshot.
Source & Registry Notes
Public provenance, Registry classification, and the latest source check for this entry, kept separate from runtime verification.
- Source repository
- falling-ts/dsh-force-compact
- Registry source
- GitHub · dsh-plugin topic
- Registry classification
- Plugin
- Source checked
- f9b42d6 · 2026-10-19
This project is independently indexed from public source information. DSH Plugin is not affiliated with DeepSeek or the plugin author. Always check the author repository before installation.
Repository Activity
- GitHub stars
- 10GitHub stars
- Forks
- 2
- Open issues
- 0
- Last commit
- 2026-10-19
- Last release
- No release detected
Related DSH Plugins
Ranked by overlapping capabilities, use cases, plugin type, categories, and DSH profile.
Persistent multi-model agent teams with dynamic DAG orchestration, run monitoring, and token insights for DeepSeek Harness.
View plugin →Use ChatGPT, Claude, Grok, GitHub Copilot, and Antigravity subscriptions as DeepSeek Harness LLM providers via OAuth.
View plugin →Session lifecycle manager for DeepSeek Harness with search, restore, idle cleanup, and diagnosis.
View plugin →Vision routing and pixel-level tools for text-only DeepSeek Harness agents with built-in free fallback.
View plugin →