安装
npx -y @deepseek-ai/dsh plugin --profile web add github:ciceroyang/dsh-trajectory此安装命令根据 GitHub 仓库地址生成,是未经验证的安装起点。
README
维护者编写的文档快照。
dsh-trajectory
Renders a DeepSeek Harness session log into a shareable HTML trajectory document — the offline, zero-dependency cousin of the official Trajectory view. "Every run leaves a trace", made portable.
Usage
node trajectory.mjs <session.jsonl.zstd>
node trajectory.mjs <sessions-dir> # newest log wins
node trajectory.mjs <sessions-dir> --all # merge every session into one chronological volume
node trajectory.mjs <log> --since 2026-08-11 --until 2026-08-17 # time-window slice (single and --all modes)
node trajectory.mjs <log> --out report.html
Output: one self-contained HTML file (inline CSS, no external assets) plus the first 16 hex chars of its SHA-256.
What the document contains
- session metadata (id / workspace / time range / turn count / token ledger)
- per-turn timeline: user asks, tool calls (argument briefs + error markers), assistant excerpts
- end-reason annotations (completed / blocked / error / …)
Technical notes
- multi-frame zstd frame scan + per-frame decode (the single-shot-decompress pitfall, algorithm ported from the official format.ts)
- zero dependencies, plain ESM; the HTML renderer is a pure function, fully unit-testable
- all output is HTML-escaped, so hostile log text cannot become script
Use cases
- delivery audit: hand the trajectory to a reviewer with a checkable SHA-256
- incident review: tool errors visible in their exact timeline position
- sharing: no DSH installation needed — open in any browser
Sibling tools
- dsh-report-studio: session → daily/weekly/handoff reports with receipts
- dsh-plugin-starter / dsh-doctor: plugin scaffold / environment doctor
项目文件与信号
以下项目是目录快照中检测到的公开仓库信号。
仓库信息
- 开发语言
- JavaScript
- 许可证
- MIT
- 最新发布
- v0.4.1
- 最后更新
- 2026年8月18日 12:07
谨慎安装
请检查源代码、权限、生命周期脚本、依赖与网络访问;不受信任的插件应先在隔离环境中测试。