DeepSeek Harness Plugins
DeepSeek Harness PluginManifest Valid

dsh-force-compact

Aggressive context compaction plugin for DeepSeek Harness agents to keep local-first prompts lean and responsive.

VisionUI & ProductivityDeveloper ToolsBrowser & WebSecurity & Policy
8GitHub Stars2ForksUpdated2026-09-16

Install

$ dsh plugin --profile web add github:falling-ts/dsh-force-compact

Plugin 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.

source-grounded
dsh-force-compact is a Cordis plugin for DeepSeek Harness designed for local-first inference workflows. It automatically compacts session context history using either an official compaction engine or a self-contained builtin fallback. By triggering at configurable token thresholds, turn ends, or manual slash commands, it summarizes older context while retaining recent tokens verbatim. It also features scoped thinking controls to disable reasoning exclusively on summarization requests, reducing token load and latency on local llama.cpp or self-hosted endpoints.

Key Capabilities

  • Automatic pre-step context compaction triggered by projected token thresholds
  • Turn-end compaction on idle agent transitions and session flush checkpoints
  • Manual trigger via the /force-compact slash command
  • Dual compaction backend support: official compaction service or builtin fallback transaction
  • Scoped thinking suppression (`reasoningEffort: 'off'` and native `reasoning_effort: "none"`) exclusively for compaction calls
  • LiveUI status badging in the web interface indicating compression states
  • Runtime configuration via settings section with live parameter updates

Useful For

  • Maintaining low context window usage when running local models like Qwen on llama.cpp
  • Reducing memory footprint and inference latency during extended multi-turn agent sessions
  • Ensuring data privacy and eliminating API token costs through fully localized context summarization

Who It Fits

  • Developers hosting local LLM inference engines (e.g., llama.cpp) with DeepSeek Harness
  • Users running long-running agent workflows constrained by hardware context limits

Documented Limitations

  • Enforces minimum value clamps on configuration (e.g., autoThresholdTokens >= 32000, retainLatestTokens >= 8000)
  • The /force-compact command registers lazily after the first model step rather than immediately at instance boot

DSH Compatibility

Version-specific runtime evidence collected by DSH Plugin. A missing result means we have not tested that combination yet.

Not tested yetNo runtime compatibility tests have been published yet.

Security Signals

Objective signals discovered from package metadata and source inspection. These are not a guarantee that a plugin is safe.

No automated security signals have been published yet.

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
15a5645 · 2026-09-20

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
8GitHub stars
Forks
2
Open issues
0
Last commit
2026-09-16
Last release
No release detected
For maintainers

Maintaining this plugin?

This listing is generated from public repository data. If you maintain this project, you can review the information and share this listing with your users if you find it useful.

Add to README
Listed on DSHPlugin.app