dsh-cybernetics
Run the following command in DeepSeek Harness:
dsh plugin install boomzikazita/dsh-cybernetics
Paste the following prompt into your AI chat to install this plugin:
Install by running dsh plugin install boomzikazita/dsh-cybernetics in DeepSeek Harness; source available at https://github.com/boomzikazita/dsh-cybernetics
About this plugin
As agent tool chains grow, loss of control becomes the real bottleneck: diverging calls, oscillating results, redundant invocations, and tool-usage collapse. dsh-cybernetics applies the classical control loop from Qian Xuesen and Song Jian s engineering cybernetics to the tool-execution pipeline, giving the scheduler a real-time feedback circuit instead of after-the-fact inspection.
Core capabilities span four layers: a feed-forward rule table with whitelist and argument-regex interception blocks out-of-scope calls; an EMA filter estimates live failure rates and auto-escalates or de-escalates across fast, deep, and conservative tiers; sliding-window detectors flag sustained divergence and high-frequency oscillation, injecting soft warnings; and an information-entropy monitor exposes tool-usage monoculture, backed by a 0-100 health score and structured failure attribution (errorType plus truncated errMsg) for weekly post-mortems.
Built for engineers running DeepSeek Harness multi-tool agent workflows who need runtime stability telemetry and adaptive throttling without modifying business code. The pure-function core (core.mjs) is unit-testable and reusable in external projects.
Use Cases
- Agent workflows with frequent tool-call failures, oscillation, or redundancy need runtime auto-throttling and alerts
- Weekly post-mortems require structured failure-root-cause distributions and tool-usage diversity metrics instead of raw log digging
- Add a classical control-loop safety valve to the tool scheduler without touching business code
Best For
- Engineers running multi-tool agent workflows on DeepSeek Harness
- Secondary developers who want unit-testable, reusable, and tunable control laws
- Platform architects focused on agent runtime stability and observability
Related Plugins
A method pack that makes AI coding agents plan against your real baseline, prove completion with fresh evidence, and reduce reworks and unsafe changes.
Turns the DeepSeek Harness session into a captain that builds a durable sub-agent team, splits goals into dependency-aware tasks, and coordinates work via direct messages and a live Web UI.
Gives coding agents design judgment, letting Claude Code, Cursor, and 70+ agents generate and iterate high-quality UI, presentations, and graphics right from the terminal.
Run the Pi ecosystem's plugins on DeepSeek Harness, unmodified, via a compatibility layer that implements Pi's public extension ABI on DSH's native services.