Session cost/usage report cards for DeepSeek Harness: tokens, cache-hit rate, per-turn breakdown, cost estimate. dsh plugin.
Pending human review — the facts below come from the source repository.
Session cost/usage report cards for DeepSeek Harness: tokens, cache-hit rate, per-turn breakdown, cost estimate. dsh plugin.
Users who want this capability in DSH; check the source and docs before installing.
dsh plugin --profile web add github:yangyongzhen/dsh-session-report No supported platforms were declared.
session/event 事件流，turn/end 后防抖 debounceMs 生成报表 - 输出 $DSH_HOME/reports/report-<id>.md（默认 ~/.dsh/reports/） - 缓存命中率按 DeepSeek 口径：cacheRead / (未缓存输入 + 缓存读 + 缓存写) - 配置单价后自动估算成本（计费输入 = 三项之和） - 进程退出时同步 flush（headless 一次性任务不丢报表） ## 安装 sh # 本地 checkout 安装 dsh plugin --profile <web|tui|headless> add file:/path/to/dsh-session-report # 或从 Git 仓库安装 dsh plugin --profile <web|tui|headless> add https://gitcode.com/qq8864/dsh-session-report.git ## 配置 yaml - id: report config: enabled: true outDir: ~/reports # 报表目录，默认 $DSH_HOME/reports debounceMs: 1000 # turn/end 后的静默等待 costPerMInput: 1.2 # 可选：每百万输入 token 价格（元），配置后估算成本 costPerMOutput: 4.0 # 可选：每百万输出 token 价格（元） 价格单位任意（元/美元/分），按"每百万 token"填写；不配置则不显示成本。 ## 报表样例 markdown # 会话成本与耗时报表 - **会话 ID**：`session-29487994-...` · **模型**：`deepseek-official/deepseek-v4-flash` - **总耗时**：2s · **估算成本**：0.0105 ## Token 消耗 | 输入 tokens（未缓存） | 6322 | | 缓存命中（读） | 1920 | | 计费输入合计 | 8242 | | 缓存命中率 | 23.3% | ## 按 Turn 拆分 | Turn | Step | 工具调用 | 输入 | 输出 | 缓存读 | 耗时 | 结果 | | 1 | 1 | 0 | 6322 | 156 | 1920 | 2s | 完成 | ## 工具调用频率 | `pwsh` | 2 | | `glob` | 1 | ## 口径说明 dsh 的 TokenUsage 计数是不相交的：inputTokens 只含未缓存输入，缓存命中/写入分开报。因此： - 计费输入 = inputTokens + cacheReadTokens + cacheWriteTokens - 缓存命中率 = cacheReadTokens / 计费输入（DeepSeek 口径，clamp 到 0–100%） - 成本估算按计费输入计算 ## 开发 sh pnpm install pnpm run build # tsc -> lib/ 端到端验证： sh dsh plugin --profile headless add file:./ dsh --profile headless --patch ./test/report-patch.yml "一句话介绍你自己" cat ~/.dsh/reports/*.md Not by stars — picked by hand, one at a time.
A visual settings dashboard for DeepSeek Harness: every registered settings namespace — including third-party plugin ones the official UI never covers — rendered as editable, schema-driven forms. Early, but the right direction: stop hand-editing YAML.
A DIY trajectory visualizer for DeepSeek Harness: watch the agent’s plan and execution unfold as a graph instead of scrolling raw logs. Early, but it fills a real gap — understanding at a glance what the agent is doing.
Dispatch DSH work from Claude Code / Codex: spin up in-host DSH agent sessions with per-tier presets, watch native subagent progress, and borrow a multimodal bridge that lends the text-only harness vision and image generation. An orchestration plugin that turns DSH into a backend for other coding agents.