dsh-token-headroom

Context compaction — keep the live prompt small, without losing the state that matters. — a DeepSeek Harness (dsh) skill, layer headroom of a four-layer token-saving family.

Compress the live context so the prompt stays small. Backs onto the dsh-headroom engine (compaction backend + CCR retrieve); this skill is the strategy layer, not a re-implementation.

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

Install

dsh plugin --profile web add dsh-token-headroom

Then invoke the skill: dsh-token-headroom.

Upstream

Based on / referencing https://github.com/WanYanTianDe/dsh-headroom. This repo is the DSH adaptation — a thin skill, not a re-implementation of the engine.

License

MIT