@jingzechen/dsh-beacon
Attention and handoff workspace for concurrent DeepSeek Harness Sessions
安装
$
npx -p @deepseek-ai/dsh dsh plugin --profile web add github:JingzeChen/dsh-beacon说明文档
阅读完整 README ↗Configuration
The published cordis.patch.yml supplies conservative defaults:
| Field | Default | Meaning |
|---|---|---|
sessionActivity.maxOpenTools | 3 | Maximum unmatched tool names projected into a Session summary |
sessionActivity.maxErrorBytes | 64 | Maximum UTF-8 bytes retained from a provider-neutral error code |
sessionBrief.automaticTriggers | turn-end, goal-blocked, turn-error | Stable checkpoints eligible for automatic generation |
sessionBrief.minMeaningfulEvents | 2 | Required activity advance after the previous accepted brief |
sessionBrief.maxBriefBytes | 4096 | Maximum complete accepted brief size |
sessionBrief.maxItemsPerField | 8 | Maximum completed, blocker, and citation items |
sessionBriefLlm.maxInputBytes | 32768 | Maximum exact JSON-framed model input |
sessionBriefLlm.maxOutputTokens | 4096 | Auxiliary output token cap |
sessionBriefLlm.timeoutMs | 30000 | End-to-end auxiliary request deadline |
A profile patch can replace the complete dsh-beacon row to change these values. DSH patch rows replace config rather than deep-merging it, so restate every required field.
To pin generated briefs to a dedicated route, add both provider and model under sessionBriefLlm. If both are omitted, Beacon uses the latest eligible route recorded for the Session. Brief generation remains unavailable when neither route exists; deterministic activity and context continue to work.