dsh-token-rtk

Input redundancy pruning — drop repeated/redundant tokens before they reach the model. — a DeepSeek Harness (dsh) skill, layer rtk of a four-layer token-saving family.

Trim verbose logs, duplicated warnings, whole-file pastes, and noise around the real signal so you pay for the signal you need, not the boilerplate around it. (RTK = Rust Token Killer strategy.)

Family: dsh-token-skills groups all four layers — handoff, rtk, headroom, caveman.

Install

dsh plugin --profile web add dsh-token-rtk

Then invoke the skill: dsh-token-rtk.

Upstream

Based on / referencing https://mintlify.wiki/rtk-ai/rtk/faq. This repo is the DSH adaptation — a thin skill, not a re-implementation of the engine.

License

MIT