ArtificialNotImbecile/dsh-context-taxonomy--packages-dsh-context-taxonomy ↗★ 1
@artificialnotimbecile/dsh-context-taxonomy
Logical-call context taxonomy for the DeepSeek Harness Web profile
AI 분석
对 DSH 的逻辑调用进行结构化脱敏捕获、压缩存储与生命周期管理。适合需要对 Agent 调用历史进行审计、分析或合规性存档的专业用户。
설치
$
npx -p @deepseek-ai/dsh dsh plugin --profile web add github:ArtificialNotImbecile/dsh-context-taxonomy#65ef5e906b654fbb03d0320de450fb2fcc9e0db2&path:packages/dsh-context-taxonomyConfiguration
The bundle patch supplies these explicit defaults:
| Field | Default | Meaning |
|---|---|---|
root | $DSH_HOME/context-taxonomy | Absolute private sidecar directory. |
captureContent | sanitized | Store sanitized logical JSON; structure-only stores summaries only. |
retentionDays | 30 | Maximum settled-capture age. |
maxCapturesPerSession | 200 | Maximum settled rows per Session lifecycle. |
maxStoredBytes | 536870912 | Global compressed-blob budget. |
maxCaptureBytes | 16777216 | Maximum sanitized bytes for one blob. Larger calls keep a summary with omitted-size-limit. |
maxPendingCaptures | 64 | Background capture admission cap. Full queues omit new captures without changing the model stream. |
extraRedactKeyPatterns | [] | Additional case-insensitive Unicode regular expressions for object keys. |
The logical-request and RPC schemas use formatVersion: 1, taxonomyVersion: 1, and source: "dsh-logical-call". Captures are fenced by Session id, creation time, and working directory so a reused id cannot read an older lifecycle's sidecar.