benchloop-client 0.2.20__tar.gz

benchloop_client-0.2.20/.gitignore

@@ -0,0 +1,15 @@
+# Python-generated files
+__pycache__/
+*.py[oc]
+build/
+dist/
+wheels/
+.omc/
+.omx/
+*.egg-info
+
+# Virtual environments
+.venv
+
+# IDE files
+.idea/

benchloop_client-0.2.20/.python-version

@@ -0,0 +1 @@
+3.12

benchloop_client-0.2.20/PKG-INFO

@@ -0,0 +1,144 @@
+Metadata-Version: 2.4
+Name: benchloop_client
+Version: 0.2.20
+Summary: Benchloop CLI — command line tool for benchmark analysis data (metrics, trajectory, reports, artifacts).
+Requires-Python: >=3.10
+Requires-Dist: httpx>=0.25.2
+Requires-Dist: tomli>=1.1.0; python_version < '3.13'
+Requires-Dist: typer>=0.9.4
+Provides-Extra: dev
+Requires-Dist: pytest>=8.0; extra == 'dev'
+Description-Content-Type: text/markdown
+
+# benchloop-client (`bl`, `bmcli`)
+
+Command line tool for fetching and downloading benchmark analysis data
+(run metrics, trajectory, pre-analysis reports, task/job lists, and artifacts)
+by `group_id` / `job_id`.
+
+> PyPI package name is `benchloop_client`; the installed commands are `bl` and `bmcli`.
+
+## Install package
+
+```
+pip install benchloop_client
+
+bl --api-key $BENCH_LOOP_API_KEY overview get -j $JOB_ID
+bl --api-key $BENCH_LOOP_API_KEY metrics get -j $JOB_ID
+
+# Option 2: config ~/.benchloop/config.toml by cli, and run bl without --api-key option:
+bl config set --api-key $BENCHLOOP_API_KEY --host https://benchmark-analysis.alibaba-inc.com
+
+bl overview get -j $JOB_ID
+bl metrics get -j $JOB_ID
+
+```
+
+The CLI checks PyPI for a newer release in the background at most once every
+24 hours. Update notices include the current/latest versions and the upgrade
+command. Notices are written only to stderr, so JSON output remains pipeable.
+
+## Install and run from source
+
+```bash
+# with uv (recommended)
+uv sync
+uv run bl --help
+
+# or editable install with pip
+python -m venv .venv && source .venv/bin/activate
+pip install -e .
+bl --help
+```
+
+## Auth & config
+
+Authentication uses a personal api key sent as the `X-API-Key` header.
+
+Precedence: CLI flag > env var > `~/.benchloop/config.toml` > default.
+
+- Env: `BENCHLOOP_API_KEY` (or `BENCHMARK_API_KEY`), `BENCHLOOP_HOST`, `BENCHLOOP_OUTPUT_DIR`
+- Default host (prod): `https://benchmark-analysis.alibaba-inc.com`
+- Pre host: `https://pre-benchmark-analysis.alibaba-inc.com`
+
+```bash
+bl config set --api-key $BENCHLOOP_API_KEY --host https://benchmark-analysis.alibaba-inc.com
+```
+
+## Usage
+
+Two-segment `bl <resource> <action>` form. `get` prints a JSON array to
+stdout (pipeable to `jq`); `download` writes the standard file layout.
+
+```bash
+bl metrics get -j J123 | jq '.[0]'
+bl metrics download -g G1 --scope all -o ./out
+bl trajectory download -g G1,G2 -o ./out
+bl trajectory download --task-id TASK_ID -o ./out
+bl report download -j J123 -o ./out
+bl report get -g G123
+bl report download -g G123 -o ./out
+bl jobs get -g G1 --all          # get supports --all (auto-paginate)
+bl jobs download -g G1           # download is always full
+bl artifacts download -g G1 -o ./out
+bl pull -g G1 --with metrics,trajectory,jobs,artifacts -o ./out
+```
+
+Group downloads always auto-paginate and include every Job in the Group. In
+particular, `trajectory`, detailed `metrics`, `report`, and `artifacts` enumerate
+the complete Job list instead of relying on the backend Group endpoints that
+return at most 1000 Jobs. Existing aggregate files keep their original paths:
+
+```text
+out/G1/trajectory.json
+out/G1/detail_metrics.json
+out/G1_report.md
+out/G1/J123_report.md
+out/G1/J123/<artifact-name>
+```
+
+`report download -g` writes both the Group summary report and each Job report.
+`pull -g` uses the same complete download behavior for every selected resource.
+Use `--skip-existing` to resume file-based downloads after an interruption.
+
+IDs accept repeated flags, comma-separated values, and `-` to read from stdin.
+`-g` and `-j` are mutually exclusive in a single command. For trajectory
+downloads, `--task-id` is a third mutually exclusive input mode.
+
+### Download filtered trajectories by task ID
+
+The benchmark analysis Jobs page can turn the current filters into a trajectory
+download task. Copy the command shown by the page and run it locally:
+
+```bash
+bl trajectory download --task-id TASK_ID -o ./out
+```
+
+The CLI fetches the task's complete job ID list, downloads job trajectories with
+fixed concurrency, and prints count/ETA progress to stderr. It creates
+`./out/TASK_ID_trajectories.zip`, containing one `<jobId>_trajectory.json` per
+successful job and a `manifest.json` with total, success, and failure details.
+If only some jobs fail, the successful files are still archived and the command
+exits with code `1`. If every job fails, no empty ZIP is created.
+
+## Exit codes
+
+| code | meaning |
+| ---- | ------- |
+| 0 | all success |
+| 1 | partial failure (some ids failed) |
+| 2 | usage / parameter error |
+| 3 | auth failure (HTTP 401) |
+| 4 | network / timeout |
+
+stdout carries data only; logs and progress go to stderr. `--quiet` suppresses
+non-error logs, `--verbose` adds debug logs.
+
+
+## Release new package
+
+```
+uv build
+twine check dist/benchloop_client-0.1.7-py3-none-any.whl
+twine upload dist/benchloop_client-0.1.7-py3-none-any.whl --verbose
+```

benchloop_client-0.2.20/README.md

@@ -0,0 +1,132 @@
+# benchloop-client (`bl`, `bmcli`)
+
+Command line tool for fetching and downloading benchmark analysis data
+(run metrics, trajectory, pre-analysis reports, task/job lists, and artifacts)
+by `group_id` / `job_id`.
+
+> PyPI package name is `benchloop_client`; the installed commands are `bl` and `bmcli`.
+
+## Install package
+
+```
+pip install benchloop_client
+
+bl --api-key $BENCH_LOOP_API_KEY overview get -j $JOB_ID
+bl --api-key $BENCH_LOOP_API_KEY metrics get -j $JOB_ID
+
+# Option 2: config ~/.benchloop/config.toml by cli, and run bl without --api-key option:
+bl config set --api-key $BENCHLOOP_API_KEY --host https://benchmark-analysis.alibaba-inc.com
+
+bl overview get -j $JOB_ID
+bl metrics get -j $JOB_ID
+
+```
+
+The CLI checks PyPI for a newer release in the background at most once every
+24 hours. Update notices include the current/latest versions and the upgrade
+command. Notices are written only to stderr, so JSON output remains pipeable.
+
+## Install and run from source
+
+```bash
+# with uv (recommended)
+uv sync
+uv run bl --help
+
+# or editable install with pip
+python -m venv .venv && source .venv/bin/activate
+pip install -e .
+bl --help
+```
+
+## Auth & config
+
+Authentication uses a personal api key sent as the `X-API-Key` header.
+
+Precedence: CLI flag > env var > `~/.benchloop/config.toml` > default.
+
+- Env: `BENCHLOOP_API_KEY` (or `BENCHMARK_API_KEY`), `BENCHLOOP_HOST`, `BENCHLOOP_OUTPUT_DIR`
+- Default host (prod): `https://benchmark-analysis.alibaba-inc.com`
+- Pre host: `https://pre-benchmark-analysis.alibaba-inc.com`
+
+```bash
+bl config set --api-key $BENCHLOOP_API_KEY --host https://benchmark-analysis.alibaba-inc.com
+```
+
+## Usage
+
+Two-segment `bl <resource> <action>` form. `get` prints a JSON array to
+stdout (pipeable to `jq`); `download` writes the standard file layout.
+
+```bash
+bl metrics get -j J123 | jq '.[0]'
+bl metrics download -g G1 --scope all -o ./out
+bl trajectory download -g G1,G2 -o ./out
+bl trajectory download --task-id TASK_ID -o ./out
+bl report download -j J123 -o ./out
+bl report get -g G123
+bl report download -g G123 -o ./out
+bl jobs get -g G1 --all          # get supports --all (auto-paginate)
+bl jobs download -g G1           # download is always full
+bl artifacts download -g G1 -o ./out
+bl pull -g G1 --with metrics,trajectory,jobs,artifacts -o ./out
+```
+
+Group downloads always auto-paginate and include every Job in the Group. In
+particular, `trajectory`, detailed `metrics`, `report`, and `artifacts` enumerate
+the complete Job list instead of relying on the backend Group endpoints that
+return at most 1000 Jobs. Existing aggregate files keep their original paths:
+
+```text
+out/G1/trajectory.json
+out/G1/detail_metrics.json
+out/G1_report.md
+out/G1/J123_report.md
+out/G1/J123/<artifact-name>
+```
+
+`report download -g` writes both the Group summary report and each Job report.
+`pull -g` uses the same complete download behavior for every selected resource.
+Use `--skip-existing` to resume file-based downloads after an interruption.
+
+IDs accept repeated flags, comma-separated values, and `-` to read from stdin.
+`-g` and `-j` are mutually exclusive in a single command. For trajectory
+downloads, `--task-id` is a third mutually exclusive input mode.
+
+### Download filtered trajectories by task ID
+
+The benchmark analysis Jobs page can turn the current filters into a trajectory
+download task. Copy the command shown by the page and run it locally:
+
+```bash
+bl trajectory download --task-id TASK_ID -o ./out
+```
+
+The CLI fetches the task's complete job ID list, downloads job trajectories with
+fixed concurrency, and prints count/ETA progress to stderr. It creates
+`./out/TASK_ID_trajectories.zip`, containing one `<jobId>_trajectory.json` per
+successful job and a `manifest.json` with total, success, and failure details.
+If only some jobs fail, the successful files are still archived and the command
+exits with code `1`. If every job fails, no empty ZIP is created.
+
+## Exit codes
+
+| code | meaning |
+| ---- | ------- |
+| 0 | all success |
+| 1 | partial failure (some ids failed) |
+| 2 | usage / parameter error |
+| 3 | auth failure (HTTP 401) |
+| 4 | network / timeout |
+
+stdout carries data only; logs and progress go to stderr. `--quiet` suppresses
+non-error logs, `--verbose` adds debug logs.
+
+
+## Release new package
+
+```
+uv build
+twine check dist/benchloop_client-0.1.7-py3-none-any.whl
+twine upload dist/benchloop_client-0.1.7-py3-none-any.whl --verbose
+```

benchloop_client-0.2.20/benchloop-cli-design.md

@@ -0,0 +1,322 @@
+# Benchloop CLI (`bl`, `bmcli`) 命令行设计
+
+> 状态：设计稿，待确认。实现将放在独立 Python 模块 `aliyun.bmcli`（主命令 `bl`，兼容别名 `bmcli`）。
+> 后端需扩展 `/openapi/**` 接口（个人 api_key 鉴权），尽量复用既有 `/openapi`。
+
+---
+
+## 1. 目标与范围
+
+提供给算法同学按 **单个或多个 group_id / job_id** 拉取基准评测数据的命令行工具：
+
+- `run_metrics`（查询 / 下载）— Group(Overall/Detail) & Job
+- `trajectory`（查询 / 下载）— Group & Job
+- 预分析报告 report（查询 / 下载）— Group & Job
+- Group / Task / Job 运行统计 overview（仅查询）
+- Task 列表 / Job 列表（查询 / 下载）— Group
+- Artifacts 产物（仅下载）— Group & Job
+
+**鉴权**：个人 api_key（HTTP 头 `X-API-Key`），对应后端 `mcp_auth_tokens` → `work_no`。
+
+**约定**：所有 `查询(get)` 返回 **JSON 数组**；即使只有一条数据，也包在数组里。
+
+---
+
+## 2. 命令总体结构
+
+统一采用 `名词(资源) + 动词` 的两段式：
+
+```
+bl <resource> <action> [ids...] [options]
+```
+
+- `action = get`     → 查询，结果打印到 stdout（JSON 数组）
+- `action = download`→ 下载，按规定的文件布局写入磁盘
+
+按规格天然区分：`overview` 只有 `get`，`artifacts` 只有 `download`，其余资源两者都有。
+
+| resource     | get | download | 适用维度        |
+| ------------ | :-: | :------: | ----------- |
+| `metrics`    |  ✓  |    ✓     | Group / Job |
+| `trajectory` |  ✓  |    ✓     | Group / Job |
+| `report`     |  ✓  |    ✓     | Group / Job |
+| `overview`   |  ✓  |    —     | Group / Job |
+| `tasks`      |  ✓  |    ✓     | Group       |
+| `jobs`       |  ✓  |    ✓     | Group       |
+| `artifacts`  |  —  |    ✓     | Group / Job |
+
+> 设计取舍：`get/download` 拆成子命令，而非用 `-o` 有无来切换——这样能精确表达「某资源只可查/只可下」，`bl <resource> --help` 也更清晰。
+
+---
+
+## 3. ID 入参约定
+
+group / job 用显式互斥语义的旗标，**可重复、可逗号分隔、可从 stdin/文件**：
+
+```
+-g, --group <id>      Group ID，可重复或逗号分隔： -g G1 -g G2 / -g G1,G2
+-j, --job   <id>      Job ID，  可重复或逗号分隔
+-g -                  从 stdin 读取（支持管道）
+```
+
+- 一条命令里 `-g` 与 `-j` 不同时出现（如 `metrics get -g G1 -j J9`）。
+- 各资源对维度的支持见上表；传入不支持的维度时报参数错误。
+
+---
+
+## 4. 全局选项（所有命令通用）
+
+```
+--api-key KEY         个人 api_key（优先级最高）
+--host URL        覆盖后端 base URL
+-o, --output-dir DIR  download 输出根目录（默认当前目录 .）
+    --pretty / --no-pretty  JSON 是否缩进（默认 pretty）
+    --timeout SEC     单请求超时（默认 300）
+    --retry N         失败重试次数（默认 2，指数退避）
+-v, --verbose         调试日志到 stderr
+-q, --quiet           只输出错误
+    --config PATH     指定配置文件路径
+    --version / --help
+```
+
+**配置与优先级**：`命令行旗标 > 环境变量 > 配置文件 > 默认值`
+
+- 环境变量：`bmcli_API_KEY`、`bmcli_HOST`、`bmcli_OUTPUT_DIR`
+- 配置文件：`~/.benchloop/config.toml`
+
+```toml
+# ~/.benchloop/config.toml
+[profile]
+# 线上: https://benchmark-analysis.alibaba-inc.com  预发: https://pre-benchmark-analysis.alibaba-inc.com
+host    = "https://benchmark-analysis.alibaba-inc.com"
+api_key = "xxxxxxxx"
+```
+
+---
+
+## 5. 各命令详细设计
+
+### 5.1 run_metrics — `bl metrics`
+
+```
+bl metrics get      [-g GroupID...] [-j JobID...] [--scope overall|detail|all]
+bl metrics download [-g GroupID...] [-j JobID...] [--scope overall|detail|all] [-o DIR]
+```
+
+- `--scope` 仅对 Group 生效（默认 `all`，即 overall + detail）：
+  - `overall` → Group 汇总指标
+  - `detail`  → Group 下每个 Job 的 run_metrics 聚合
+- Job 维度忽略 `--scope`，直接取该 Job 的 run_metrics。
+
+**下载文件布局**
+
+| 维度          | 文件                                |
+| ----------- | --------------------------------- |
+| Group overall | `{group_id}/overall_metrics.json` |
+| Group detail  | `{group_id}/detail_metrics.json`  |
+| Job           | `{job_id}_run_metrics.json`       |
+
+### 5.2 trajectory — `bl trajectory`
+
+```
+bl trajectory get      [-g GroupID...] [-j JobID...]
+bl trajectory download [-g GroupID...] [-j JobID...] [-o DIR]
+```
+
+**下载文件布局**
+
+| 维度    | 文件                          |
+| ----- | --------------------------- |
+| Group | `{group_id}/trajectory.json` |
+| Job   | `{job_id}_trajectory.json`  |
+
+### 5.3 预分析报告 — `bl report`
+
+```
+bl report get      [-g GroupID...] [-j JobID...]
+bl report download [-g GroupID...] [-j JobID...] [-o DIR]
+```
+
+- `get` 时返回 JSON 数组。Job 报告包含 `fail_reason`、`sub_fail_reason`；Group 报告还包含组内 Job 的两级错误分布统计。
+
+**下载文件布局**
+
+| 维度  | 文件                      |
+| ----- | ------------------------- |
+| Group | `{group_id}_report.md`    |
+| Job   | `{job_id}_report.md`      |
+
+### 5.4 运行统计 overview — `bl overview`（仅查询）
+
+```
+bl overview get -g GroupID... --scope summary   # Group Overview（eval_summary）
+bl overview get -g GroupID... --scope tasks      # Group Task Overview（组内 task 维度统计，eval_tasks）
+bl overview get -g GroupID...                     # Group：summary + tasks（--scope all，默认）
+bl overview get -j JobID...                       # Job Overview（jobs/{id} + jobs/stats）
+```
+
+- 仅 `get`，不提供下载；维度由 `-g`/`-j` 决定；Group 维度的 `--scope summary|tasks|all`（默认 all）与 `metrics get` 的 `--scope` 保持一致。输出 JSON 数组。
+
+### 5.5 Task 列表 — `bl tasks`（Group）
+
+```
+bl tasks get      -g GroupID... [--page N] [--page-size M] [--all] [--search ...] [--status ...]
+bl tasks download -g GroupID... [-o DIR]
+```
+
+- `--all`：自动翻页拉全量（download 默认全量）。
+- 文件：`{group_id}/task_list.json`
+
+### 5.6 Job 列表 — `bl jobs`（Group）
+
+```
+bl jobs get      -g GroupID... [--page N] [--page-size M] [--all]
+                   [--template ...] [--status ...] [--instance-id ...]
+bl jobs download -g GroupID... [-o DIR]
+```
+
+- 文件：`{group_id}/job_list.json`
+
+### 5.7 Artifacts 产物 — `bl artifacts`（仅下载）
+
+```
+bl artifacts download [-g GroupID...] [-j JobID...] [-o DIR]
+                        [--include GLOB] [--exclude GLOB] [--list-only]
+```
+
+- Job：下载该 Job 全部产物到 `{job_id}/` 目录。
+- Group：下载组内所有 Job 产物到 `{group_id}/` 目录（接入 Lindorm）。
+- `--list-only`：只列出产物清单不下载；`--include/--exclude` 按通配过滤。
+- 后端返回 OSS 预签名 URL，CLI 负责并发下载与落盘。
+
+**下载文件布局**
+
+| 维度    | 路径            |
+| ----- | ------------- |
+| Job   | `{job_id}/`   |
+| Group | `{group_id}/` |
+
+### 5.8（可选）一键打包 — `bl pull`
+
+便捷聚合命令：对给定 group/job 一次性拉取 metrics + trajectory + lists(+report/artifacts) 到上述布局。
+
+```
+bl pull -g GroupID... [-j JobID...] [--with metrics,trajectory,report,artifacts,tasks,jobs] [-o DIR]
+```
+
+---
+
+## 6. 输出文件布局总览（download 根目录下）
+
+```
+<output-dir>/
+├── {group_id}/
+│   ├── overall_metrics.json
+│   ├── detail_metrics.json
+│   ├── trajectory.json
+│   ├── task_list.json
+│   ├── job_list.json
+│   └── <artifacts...>            # group 产物
+├── {job_id}_run_metrics.json
+├── {job_id}_trajectory.json
+├── {job_id}_report.md
+└── {job_id}/                     # job 产物
+    └── <artifacts...>
+```
+
+---
+
+## 7. 后端 `/openapi` 接口对照（复用 vs 待补充）
+
+> 鉴权统一为 `X-API-Key`（既有机制：`mcp_auth_tokens` → `work_no`）。
+
+| CLI 命令                         | 复用既有 `/openapi`                              | 需新增 `/openapi`                                  |
+| ------------------------------ | -------------------------------------------- | ---------------------------------------------- |
+| `metrics get/download -j`      | `GET /openapi/metrics/{jobId}` ✅             | —                                              |
+| `metrics ... -g --scope overall` | `GET /openapi/groups/{groupId}/eval_summary` ✅ 或 `GET /openapi/metrics?group_id=` | （二选一，确认 overall 字段口径）             |
+| `metrics ... -g --scope detail`  | 可基于 `GET /openapi/jobs?group_id=` + 逐 job `metrics/{jobId}` | 建议新增 `GET /openapi/groups/{groupId}/metrics/detail` 直接聚合 |
+| `trajectory ... -j`            | —                                            | **新增** `GET /openapi/jobs/{jobId}/trajectory` |
+| `trajectory ... -g`            | —                                            | **新增** `GET /openapi/groups/{groupId}/trajectory`（或复用 `trajectory/download` 任务流） |
+| `report ... -j`                | —                                            | **新增** `GET /openapi/jobs/{jobId}/report` |
+| `report ... -g`                | —                                            | **新增** `GET /openapi/groups/{groupId}/report` |
+| `overview get -g --scope summary` | `GET /openapi/groups/{groupId}/eval_summary` ✅ | 视字段补充                                |
+| `overview get -g --scope tasks`   | `GET /openapi/groups/{groupId}/eval_tasks` ✅（聚合口径）| 可能新增统计汇总端点                 |
+| `overview get -j`                 | `GET /openapi/jobs/{jobId}` ✅ + `GET /openapi/jobs/stats` ✅ | —                           |
+| `tasks get/download -g`        | `GET /openapi/groups/{groupId}/eval_tasks` ✅ | —                                              |
+| `jobs get/download -g`         | `GET /openapi/jobs?group_id=&skip=&limit=` ✅ | —                                              |
+| `artifacts download -j`        | —                                            | **新增** `GET /openapi/jobs/{jobId}/artifacts`（返回预签名 URL 列表） |
+| `artifacts download -g`        | —                                            | **新增** `GET /openapi/groups/{groupId}/artifacts`（分页，Lindorm） |
+
+**待后端补充实现清单（汇总）**
+
+1. `GET /openapi/jobs/{jobId}/trajectory`
+2. `GET /openapi/groups/{groupId}/trajectory`
+3. `GET /openapi/jobs/{jobId}/report`
+4. `GET /openapi/groups/{groupId}/report`
+5. `GET /openapi/jobs/{jobId}/artifacts`
+6. `GET /openapi/groups/{groupId}/artifacts`
+7. `GET /openapi/groups/{groupId}/metrics/detail`（detail 聚合，可选）
+
+---
+
+## 8. 行为约定
+
+- **多 id**：使用内部固定并发数执行，结果数组合并；download 各自落盘。
+- **返回数组**：`get` 始终输出数组；单条也包成单元素数组。
+- **部分失败**：某 id 失败不阻断其它 id；汇总在结尾打印失败清单，退出码非 0。
+- **退出码**：`0` 全部成功；`1` 部分失败；`2` 参数错误；`3` 鉴权失败(401)；`4` 网络/超时。
+- **stdout/stderr 分离**：数据走 stdout（可管道给 `jq`），日志/进度走 stderr。
+- **幂等下载**：默认覆盖；`--skip-existing` 跳过已存在文件；`--list-only` 仅预览。
+
+---
+
+## 9. 打包与分发
+
+- 模块：`aliyun.bmcli`，可执行入口 `bl` 与兼容别名 `bmcli`（`[project.scripts] bl = "aliyun.bmcli.__main__:main"`，`bmcli = "aliyun.bmcli.__main__:main"`）。
+- 技术栈建议：Python 3.10+ + Typer/Click + httpx，`uv` + `hatchling` 打包（与现有 `bench_mcp` 工具链一致）。
+- 安装：`uv tool install benchloop_client` 或 `pipx install benchloop_client`。
+
+---
+
+## 10. 用法示例
+
+```bash
+# 配置一次
+bl config set --api-key $KEY --host https://benchmark-analysis.alibaba-inc.com
+
+# 查询单个 job 的 run_metrics（数组输出，可接 jq）
+bl metrics get -j J123 | jq '.[0].score'
+
+# 下载一个 group 的 overall+detail 指标
+bl metrics download -g G1 --scope all -o ./out
+
+# 多个 group 的 trajectory 下载
+bl trajectory download -g G1,G2 -o ./out
+
+# job 的预分析报告
+bl report download -j J123 -o ./out
+
+# group 报告及组内错误分布
+bl report get -g G1
+bl report download -g G1 -o ./out
+
+# group 的 job 列表（全量）
+bl jobs download -g G1 --all
+
+# 下载 group 下所有 job 产物
+bl artifacts download -g G1 -o ./out
+
+# 一键拉取
+bl pull -g G1 --with metrics,trajectory,jobs,artifacts -o ./out
+```
+
+---
+
+## 11. 其他问题
+
+1. `overall metrics` 是 Group 页面 Run_Metrics 下面显示的统计的 metrics 数据.
+2. `metrics detail`  CLI 端循环按照 group_id 分页拉取
+3. `overview task`（Group Task Overview）统计口径与现有 `eval_tasks` 的差异。
+4. trajectory（尤其 group 级，数据量大）走「同步返回」还是「异步下载任务」模式。
+5. artifacts group 级量大时是否需要断点续传 / 限速。
+6. api_key 是否复用 `mcp_auth_tokens`，还是单独签发体系。