SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillsdingtalk-minutes

dingtalk-minutes

Office & productivity

View repository

Updated 10-15 20:10

About this skill

DingTalk AI Minutes Read Skill (thin-shell wrapper)

This skill is a dws(DingTalk official Workspace CLI) minutes service'sread capability wrapper, focusing on "querying and reading AI Minutes content" —contains no write operations such as writing/modifying/recording control on the DingTalk server side(does not call update/upload/record and other write commands). All reads are performed through dws , without bypassing the CLI to call the HTTP API directly.

Note: "read-only" in this skill meansnot modifying DingTalk cloud data; however, archiving (archive), mirroring (mirror), and optional audio download are allexplicit operations that write already-read content to the user's local file system, requiring the user to actively run the corresponding script and specify a directory; this is not equivalent to unauthorized external transmission. See the sections "Local Archiving and Incremental Sync" and "Mirroring to External Folders" below for details.

Command reference (read-only type):references/01-commands.md。 Must-read for first deployment (installation/authorization/pitfalls)：references/02-setup.md。

Dependencies

System dependencies

Dependency Installation method
dws(DingTalk official Workspace CLI) macOS/Linux: download the installation script first, then execute it (see the command below; note the path is scripts/install.sh, not the root directory install.sh）
python3(used by the sync script) macOS usually includes it; if missing, brew install python
curl(used by the installation script) macOS usually includes it; if missing, brew install curl

Install dws(download to a temporary file first, then execute, to avoid curl | sh executing a remote script directly):

curl -fsSL https://raw.githubusercontent.com/DingTalk-Real-AI/dingtalk-workspace-cli/main/scripts/install.sh -o /tmp/dws-install.sh
sh /tmp/dws-install.sh

PATH configuration: dws is installed to ~/.local/binby default and needs to be added to the shell PATH. The following command will modify your ~/.zshrc(persistent shell configuration), appending only one PATH line and being idempotent (skipped if already present); if you do not want to automatically modify the configuration, you can manually add export PATH="$HOME/.local/bin:$PATH" to your shell configuration:

grep -q '.local/bin' ~/.zshrc || echo 'export PATH="$HOME/.local/bin:$PATH"' >> ~/.zshrc

Authorization and organization switch (both are indispensable)

This skill depends on three independent prerequisites; if any step is missing, reading will fail: install dws → enable the organization's "CLI Access Management" switch → authorize and log in. For detailed steps, the counterintuitive semantics of the switch, and pitfalls such as authorization must run in the background, see references/02-setup.md。

Out-of-the-box and dependency-required features

  • Out-of-the-box: This skill's core read commands (list / +detail / get *) are all calls to dws; once dws is installed and authorized, they can be used.
  • Requires local Python：scripts/sync.py(local archiving and incremental sync),minutes_extract_todos.py and other scripts require python3(standard library only, no third-party package dependencies).

Prerequisites

  1. dws is installed：dws version and can return normally (see the "Dependencies" installation instructions above).
  2. Authorized：dws auth status shows authenticated: true. For authorization use dws auth login --device（run in the background, wait for the callback to complete after scanning the QR code; see the pitfalls document).
  3. Organization switch enabled: DingTalk Open Platform → Developer Platform → More → Basic Information → CLI Access Management, when the toggle text is "Prohibit all members from using CLI", you need toturn it offOnly then is it allowed (see the pitfalls document for details).
  4. All dws commandsmust include --format json to get parseable output.

Core capabilities (read)

User intent Command
List my meeting notes dws minutes list mine [--query "<关键词>"] [--start "<ISO>"] [--end "<ISO>"]
List all meeting notes I can access (including those shared by others) dws minutes list all
List those shared with me by others dws minutes list shared
Fetch everything at once (basic information/summary/keywords/verbatim transcript/todos) dws minutes +detail --uuid <taskUuid>
Read AI summary dws minutes get summary --id <taskUuid>
Read the original speech-to-text transcript (verbatim transcript) dws minutes get transcription --id <taskUuid>
Read keywords dws minutes get keywords --id <taskUuid>
Read todos dws minutes get todos --id <taskUuid>
Read audio download URL dws minutes get audio --id <taskUuid>
Merge recent meeting note summaries python scripts/minutes_recent_summary.py --max 5
Extract todos from a certain meeting python scripts/minutes_extract_todos.py --id <taskUuid>

Standard SOP

SOP-1 Query meeting note list (query-minutes)

  1. Choose scope (iron rule)：mine=created/initiated by me;shared=shared with me by others;all=all I can access (mine∪shared). When the user says "I can access/visible/all/my meeting notes", always use all; only when they explicitly say "created by me/initiated by me" use mine。
  2. Execute：dws minutes list all|mine|shared --format json; add keywords with --query, add time with --start/--end, limit the number of items with --max <n>, paginate with --next-token <token>。
  3. Parse: from itemList[] get the real taskUuid + title + time; if there are multiple candidates, have the user confirm, do not default to taking the first one.

SOP-2 Get meeting note details (get-minute-detail)

  1. Prerequisite: first follow SOP-1 to get the target taskUuid。
  2. Execute (choose one as needed)：get summary / get transcription / get keywords / get todos / get info / get audio, all include --format json。
  3. Transcript pagination (required)：get transcription A single call returns at most about 50 segments; when the return includes nextToken , thenmustcontinue --next-token pagination to pull everything, then summarize.
  4. Parse：--id/--uuid/--task-uuid are equivalent,recommended to uniformly use --id; it is prohibited to fabricate taskUuid.

High-frequency hard constraints

  • Automatic URL extraction: when the user provides a shanji.dingtalk.com/app/transcribes/<taskUuid> -type link, automatically extract the hex string as --id; it is prohibited to use the entire URL as a parameter, and prohibited to open it in a browser.
  • Calculate time yourself: when the user says today/this week/last week/last N days/a certain date range, calculate --start/--endyourself (ISO-8601, such as 2026-05-11T00:00:00+08:00), do not ask back.
  • Server-side filtering first: when the time range and keywords can be filtered server-side, they must be put into the same list all --start --end --query, do not pull everything back and filter locally.
  • Empty list fallback: for the same range list all → remove keywords and keep time → clearly inform that there is no data. It is prohibited to fabricate meeting note content to generate minutes.
  • Fetch data before generating: before generating minutes/documents/todos, you must first list → lock onto the real taskUuid → get summary(if the original text/action items are needed, then get transcription/get todos). If the data is not obtained, stop and explain the blocker.
  • Export the original text without downgrading: When the user wants to "download/export the verbatim transcript", you must retrieve it item by item get transcription and page through to the end; do not downgrade to a summary.
  • Shell pipes are prohibited: Do not use |、head、grep、jq to truncate the output; use --format json to process in memory.

Local archiving and incremental sync (archive / sync)

This skill supports syncing DingTalk AI meeting notesto the archive/ directory within the skill, creating a local backup to avoid forgetting historical content and to only incrementally pull new meeting notes.

⚠️ Privacy and compliance notice: Verbatim transcripts, summaries, and to-dos may contain client secrets, the personal privacy of involved parties, or internal business information. Archiving/mirroring will write the above contentto the local file system—do not commit archive/ or the mirror target directory to public repositories, shared directories, or third-party sync services;archive/ and the local mirror configuration have been added by default to .gitignore. Before running, confirm that the target location is accessible only to you.

Archive structure

archive/
├── index.json                 # 同步状态：last_sync(上次同步时间) + synced_uuids + uuid_to_dir(uuid→目录名映射)
└── <YYMMDD>_<标题>/           # 目录名：日期(两位年，如 260508) + 下划线 + 听记标题
    ├── meta.json              # 结构化元数据：uuid/标题/时间/时长/分享链接/创建人/关键词列表/音频信息
    ├── transcript.md          # 语音转写逐字稿（已翻页拉全，含【发言人 N】前缀）；文件头部含关键词/AI摘要/待办概览
    ├── summary.md             # AI 生成的完整摘要（fullSummary 全文）
    ├── keywords.md            # 关键词列表
    └── todos.md               # 待办事项（含负责人，来自 get todos 详细接口）

Goal: Store as much as possible in the internal archive. All textual information that can be extracted from a single meeting note will be saved to disk—verbatim transcript, AI summary, keywords, to-dos, audio download URL, and metadata. The audio file itselfis not downloaded by default(the URL carries expiring authentication, and a single item is about 150MB); if you need a local audio backup, run python scripts/sync.py --with-audio。

Directory name example:260805_08-05 图书出版协作优化/. When there is a conflict with the same date and same title, append a short uuid suffix (e.g., 260805_xxx_3af2c1). Deduplication and incremental determination are based on uuid; the directory name is only for readability, and it is traced back via the index.json of uuid_to_dir .

Sync command

python scripts/sync.py                  # 增量同步：仅拉取 last_sync 之后的新听记
python scripts/sync.py --full           # 全量重扫（已存在 uuid 跳过，不重复拉逐字稿）
python scripts/sync.py --list-new       # 只列出本次新增标题，不拉逐字稿
python scripts/sync.py --dry-run        # 预览将执行的 dws 命令，不写文件
python scripts/sync.py --archive-dir /path/to/archive   # 指定存档目录
python scripts/sync.py --no-mirror      # 本次只存档，不自动镜像到外部文件夹

Automatic mirroring (enabled by default): When there are new archives in this run and config/mirror-target.local.json exists, automatically call after the sync is complete mirror_output.py for incremental mirroring (sha256 verification, also filling in files that previously failed to mirror); if no mirror target is configured, prompt to skip without affecting archiving. Mirror failure is also only reported—archive is the authoritative source.

Incremental principle

  1. Read archive/index.json of last_sync as the parameter of dws minutes list all --start <last_sync> ; the server returns only meeting notes after that time.
  2. Existing local synced_uuids entries are skipped to avoid repeated pulls.
  3. For each new meeting note: generate the directory name according to YYMMDD_标题 , pull get transcription(page through to pull everything) and save transcript.md, pull get summary/get todos/get keywords and save meta.json; record the uuid-to-directory-name mapping in uuid_to_dir。
  4. Update index.json: write the startTimeISO of the latest meeting note into last_sync, and merge the uuid into synced_uuids。

Usage conventions

  • On first run there is no index.json → full scan (limited by the dws list pagination; the script automatically pages through).
  • To know "where the last sync got to and what was newly added this time" → look at the last_sync output by the script and the list of newly added titles, or directly read archive/index.json。
  • The archive directory is added to .gitignoreper the AGENTS.md convention (or managed separately in a private repository), to avoid mistakenly committing verbatim transcripts to a public repository.

Mirroring to an external folder (mirror)

One-way copy of the finished meeting notes from archiveOne-way copyTo an external designated folder (such as the Obsidian / Clawd knowledge base), for manual review. The archive is the authoritative source; the mirrordoes not write back to the archiveand does not change the sync status.

sync.py When there are new items, by default it automatically mirrors(since v1.1.0) — once sync completes, the mirroring process below runs automatically, with no need to manually run the commands in this section; the commands in this section are used for manual gap-filling, changing the target, filtering by date/single item, and similar scenarios.

Usage

# 默认按 config/mirror-target.local.json 的 dest 镜像全部听记
python scripts/mirror_output.py

# 只镜像指定日期（YYMMDD）之后开始的听记
python scripts/mirror_output.py --since 260801

# 只镜像单条听记（指定 archive 内目录）
python scripts/mirror_output.py --archive "archive/260805_08-05 图书出版协作优化"

# 覆盖目标目录 / 自定义白名单
python scripts/mirror_output.py --dest /path/to/output
python scripts/mirror_output.py --items transcript,summary,todos,keywords

# 预览将复制哪些文件，不写入
python scripts/mirror_output.py --dry-run

Configuration

Copy the template to a local config and edit dest：

cp config/mirror-target.example.json config/mirror-target.local.json
# 编辑 .local.json 的 dest 字段（本机实际路径）

config/mirror-target.local.json has been .gitignore excluded (machine-specific path, not committed to the version repository). If the config file is missing and --dest is not passed, it exits with code 2 and a prompt.

Mirrored content (whitelist)

key File Description
transcript transcript.md Speech-to-text verbatim transcript (including overview)
summary summary.md Full AI summary text
todos todos.md To-do items
keywords keywords.md Keywords (optional)
meta meta.json Structured metadata (optional)

By default only mirrors transcript,summary,todos(three md files).Does not copy meta.json(unless explicitly adding --items ... ,meta), to avoid leaking structured internal data. Missing files are skipped without error (some recordings themselves have no todos/summary).

Mirror directory structure

<dest>/
├─ 260805_08-05 图书出版协作优化/
│   ├─ transcript.md
│   ├─ summary.md
│   ├─ todos.md
│   └─ .mirror-manifest.json     # 源路径 + 文件列表 + sha256，便于核对
└─ 260729_07-29 医疗损害鉴定听会/
    └─ ...

Incremental and verification

  • If the target file already exists and the sha256 matches → treated as already mirrored, skipped without overwriting (incremental).
  • Each mirror run writes under the recording subdirectory .mirror-manifest.json(source archive path, mirror time, file list + sha256), for later verification.
  • A mirror failure is only reported and does not affect the archive or sync status.

Cross-product collaboration

  • Create tasks in bulk from to-dos → switch dingtalk-todo
  • Send the summary to a colleague → switch dingtalk-chat
  • Schedule/meeting room → switch dingtalk-calendar
  • Save to disk as a document → switch dingtalk-doc

Scope boundary of this skill (thin shell)

  • ✅ Only wrapsreadcapabilities: list, summary, transcription, keywords, to-dos, audio addresses, recent merge, to-do extraction script.
  • ❌ Does not include write operations: modifying title/summary (update), full-text replacement (replace-text/+replace-batch), uploading audio (upload), recording control (record), permission management (permission), speaker matching/correction. If you need these, directly call dws minutes <cmd> or refer to DingTalk's official dws built-in dingtalk-minutes full documentation.

Dependencies

System dependency: requires installing dws(DingTalk's official Workspace CLI, version >= 1.0.15).

Python packages: no third-party dependencies (scripts underscripts/ use only the standard library).

References and acknowledgements

  • The command contract originates from DingTalk's official dingtalk-workspace-cli built-in dingtalk-minutes skill; this skill refines it into a read-only thin shell. For command reference see references/01-commands.md。