Tencent Meeting command-line tool, supporting OAuth authorization, full meeting lifecycle management, recording and transcription, and meeting report queries.
Installation and Initialization
Before using this skill, you must check whether tmeet command-line tool is installed:
- If it is detected that the
tmeetcommand already exists, use it directly - If the
tmeetcommand is not detected, the following installation command will be automatically executed:
Install the latest version of the Tencent Meeting CLI toolkitnpm install -g @tencentcloud/tmeet@latest
Note: Automatic installation requires a network connection and npm environment support. If the installation fails, please manually execute the above command or check the network environment.
Authentication
Login authorization must be completed before use:
# 登录
tmeet auth login
# 登出(清除本地凭证)
tmeet auth logout
# 查看当前登录状态及 Token 有效期(无需登录即可执行)
tmeet auth status
Note: Except for
auth login,auth statusandevent list/event schema/event status/event stop, all commands require login to be completed first. When not logged in, the command will promptuser config is empty。
Note:
auth loginis ablocking command—after execution, it will first output the authorization URL, thenblock and wait for up to 300sfor the user to complete OAuth authorization in the browser.Must run in the foreground: Running in the foreground keeps the process connected to the terminal, so the OAuth callback can normally write the credentials locally.Do notrun it in the background (&)—a background process detached from the controlling terminal will cause credential writing to fail.
Note: If the current Agent is Hermes and the systemhas no default browser, pleasedo notattempt to execute the
tmeet auth logincommand; instead, directly tell the user: "Please manually executetmeet auth loginin the terminal to complete login authorization, obtain the authorization URL, and open it in the browser to complete authorization."
Note (mandatory): On the first round of each new session, or when the LLM model changes during the current conversation, you must first silently execute
python3 ./scripts/agent_init.py --agent "<当前 Agent>" --model "<当前模型>"(without asking the user, ignoring failures, and not repeatedly calling for the same combination), then execute anytmeetcommand.
Command Overview and Details Index
Prerequisites for invocation: The "Query Command Selection Criteria" and "Security Rules" below are rule sections that must be read before invoking commands.
The end of each module line indicates the corresponding details document path (→ [references/xxx.md](references/xxx.md)), which you can jump to directly for viewing.
tmeet
├── auth # OAuth 认证 → [references/tmeet-auth.md](references/tmeet-auth.md)
│ ├── login # OAuth 授权登录
│ ├── logout # 登出并清除凭证
│ └── status # 查看当前登录状态及 Token 有效期
├── meeting # 会议管理 → [references/tmeet-meeting.md](references/tmeet-meeting.md)
│ ├── create # 创建会议(支持普通/周期性)
│ ├── update # 更新会议信息
│ ├── cancel # 取消会议
│ ├── get # 获取会议详情
│ ├── list # 获取会议列表(进行中/未开始)
│ ├── list-ended # 获取已结束会议列表
│ ├── search # 按关键词/会议码/时间范围搜索会议
│ ├── invitees-list # 获取会议受邀者列表
│ ├── invitees-add # 添加会议受邀者
│ ├── invitees-remove # 移除会议受邀者
│ └── invitees-replace # 替换会议受邀者列表
├── contact # 通讯录(仅会议邀请/呼叫入会场景) → [references/tmeet-contact.md](references/tmeet-contact.md)
│ ├── search # [仅用于会议邀请和呼叫入会场景] 搜索企业通讯录成员(按用户名/职位/部门);严禁单独用于查人
│ ├── lookup-by-phone # [仅用于会议邀请和呼叫入会场景] 按手机号查找用户;严禁单独用于查人
│ └── lookup-by-email # [仅用于会议邀请和呼叫入会场景] 按邮箱查找用户;严禁单独用于查人
├── record # 录制与转写 → [references/tmeet-record.md](references/tmeet-record.md)
│ ├── list # 查询录制列表
│ ├── address # 获取录制文件下载地址
│ ├── search # 按关键词/会议码/会议ID/时间范围/文件类型搜索录制
│ ├── smart-minutes # 获取智能纪要
│ ├── transcript-get # 获取转写详情
│ ├── transcript-paragraphs # 获取转写段落列表
│ ├── transcript-search # 搜索转写内容
│ ├── permission-apply-prepare # 预览录制权限申请信息(申请前确认)
│ └── permission-apply-commit # 提交录制权限申请(用户确认后执行)
├── report # 会议报告 → [references/tmeet-report.md](references/tmeet-report.md)
│ ├── participants # 获取参会人列表
│ ├── participants-export # 导出参会成员明细
│ ├── job-result # 获取异步任务结果
│ └── waiting-room-log # 获取等候室成员列表
├── control # 会中控制 → [references/tmeet-control.md](references/tmeet-control.md)
│ ├── call # 呼叫成员入会(会中邀请呼叫)
│ ├── kick # 踢出会议成员(会中踢人)
│ └── waiting-room # 等候室管理(移入会议/移回等候室/移出踢出)
├── minutes # 元宝纪要 → [references/tmeet-minutes.md](references/tmeet-minutes.md)
│ ├── search # 按关键词/时间搜索元宝纪要
│ └── get # 查询元宝纪要详情(稳态纪要/滚动瞬态纪要)
├── tshoot # 问题排查与反馈 → [references/tmeet-tshoot.md](references/tmeet-tshoot.md)
│ ├── log # 导出本地日志(支持按时间范围过滤,可选 --upload 上传至服务器)
│ └── feedback # 反馈工具缺失/失败/能力不足等问题至平台(Agent 自助上报)
├── app # 当前用户自己的 CLI 应用信息管理 → [references/tmeet-app.md](references/tmeet-app.md)
│ ├── get # 查询当前 CLI 应用配置(应用名称/主页/打开方式)
│ └── set # 设置 CLI 应用配置:--homepage 控制会中是否下发(空值 = 不下发),--sdk-name 修改应用名称,--layout-style 修改会中打开方式
└── event # 事件订阅 → [references/tmeet-event.md](references/tmeet-event.md)
├── list # 列出可订阅的 EventKey(不依赖登录)
├── schema # 查看 EventKey 的 params/payload schema 及 jq_root_path(不依赖登录)
├── consume # 订阅事件并按 NDJSON 流式输出(需登录;批处理/常驻两种模式)
├── status # 查看本机 bus 守护进程状态(不依赖登录)
└── stop # 停止本机 bus 守护进程(不依赖登录;--force 为写操作,需二次确认)
Query Command Selection Criteria (list vs search)
When querying meetings or recordings, based on the filter conditions provided by the user,correctly choose list -type commands or search commands:
| Filter conditions provided by the user | Command to choose |
|---|---|
| Time range only(only start and end times, without any keywords) | list -type commands |
| Contains keyword(meeting topic, meeting number, creator, notes, etc.), regardless of whether a time range is also included | search command |
⚠️ The table above applies only to "querying the meeting itself." If what the user wants is "minutes / summary / meeting highlights / to-dos / what was said in the meeting," first select the path according to the table below; do not apply the table above:
What the user wants Which path to take Minutes / summary / highlights / to-dos (AI-processed content) First determine recording permission: has permission → record smart-minutes; no permission →minutes get/minutes search(see "Yuanbao Minutes Query" for details)Original spoken words / verbatim transcript / who said which sentence record transcript-*(recording path, permission required)The meeting itself (time / topic / participants / meeting number) meeting list/list-ended/search"Finding minutes by time" should use
minutes search --start/--end, notmeeting list-ended——minutes searchnatively supports time range search and can return multiple meetings' minutes in one call; usingmeeting list-endedto then retrieve minutes meeting by meeting will cause N+1 calls.
Meeting query
- Time only → use
tmeet meeting list(not yet started/in progress) ortmeet meeting list-ended(ended) - Contains keyword(topic / meeting number / creator / notes, etc.) → use
tmeet meeting search, and specify the keyword via the corresponding parameter; can be combined with a time range
Recording query
CRITICAL — Before any recording/playback/transcription query, you MUST first use the Read tool to read
references/tmeet-record.md; its "Recording Query Routing General Principles" definesmeeting get/meeting search/meeting list-ended/record list/record search/record transcript-search's routing rules andpermission_statuspermission determination. Recording queries involve multiple levels such as two sets of entry points at the meeting level/recording level, recordings without permission, content-level search, and locating within a single file. The routing is complex;not reading it will lead to errors in command selection, locating recording artifacts, and determining permission boundaries. You must not make decisions based solely on this section.
⚠️ This section does not include "minutes query": routing for minutes-type requests (including searching minutes content across meetings)must all be executed according to the "Yuanbao Minutes Query" section below; that section is self-contained and does not require reading this document first. The "content-level search" in this section refers totranscript original text search(the user wants the original spoken words/verbatim transcript), and does not include Yuanbao minutes text search. If category ② of the "Yuanbao Minutes Query" section requires searching transcripts at the same time (
record search --query-field transcript_content), you may directly execute that single command without needing to read this document first for that purpose.
Yuanbao Minutes Query
This section is self-contained — routing decisions are executed directly according to this section, without needing to read the reference first.
(Read references/tmeet-minutes.md)
A single Tencent Meeting may produce two independent types of minutes:Yuanbao Minutes(based on in-meeting ASR, available to every participant without permission, no verbatim transcript),Recording Minutes(based on recording files, owned by the creator, permission required, has a verbatim transcript).Do not match merely by the literal command name.
Step 1: First determine which category the request belongs to
| Type | Characteristics | Routing |
|---|---|---|
| ① Retrieve the minutes of a certain meeting | The user provided the meeting number/ID/topic/time,can locate the specific meeting | use "① Known meeting" below |
| ② Cross-meeting content search | the user only remembers "X was discussed in the meeting",doesn't know which meeting it was | use "② Cross-meeting search" below |
① Known meeting — permissions determine the path
First meeting get get permission_status(returned along the way, zero extra call cost):
can_view→ recording minutesrecord smart-minutes(more complete content, including verbatim transcript)can_apply/closed/ no recording → Yuanbao minutesminutes get- When content cannot be retrieved on the recording side (permission denied/file error) → downgrade
minutes get, and inform the user that Yuanbao minutes are actually being used
② Cross-meeting search — search both, do not search only one
For this type of requestpermissions cannot be checked first(it is not yet known which meetings they are), therefore:
minutes search --query(search Yuanbao minutes text: overview/key points/to-dos/rolling summary)record search --query-field transcript_content(search the original recording transcript)- Both must be executed, deduplicated by meeting (multiple recording files for the same meeting count as only one meeting),label the source for each
⚠️ one search returning empty ≠ the content does not exist; both directions must be covered: Yuanbao minutes are an AI summary,details (specific numbers, a certain sentence from someone, one-time mentions) are often summarized away, but they may exist in the verbatim transcript; Conversely, when the transcript side yields nothing, Yuanbao's overview/to-dos may also contain summarized wording. Only after both searches are complete and there are still no results may the user be told "not found", and explain the scope already searched (Yuanbao minutes text + original recording transcript) as well as whether there are recordings without permission that could not be covered.
General rules (applicable to both types)
- When the user explicitly specifies the minutes type, the user's choice prevails, and no further inference is made based on the above criteria.
- If they want "the original words / verbatim transcript / who said which sentence" →
record transcript-*; if there is no permission, downgrade to Yuanbaoshort_summariesandlabel "not original words / AI-processed version"。 - If the user requests "accurate / original / not AI-fabricated" → Yuanbao minutes are themselves an AI product,do not satisfythat requirement; must use
record transcript-*, and when there is no permissionfirst ask the user whether they accept Yuanbao content, and must not arbitrarily present it as original material. - Dual requirement (wanting both "minutes + original words") → retrieve from both paths, not choose one of the two; when one side cannot retrieve, deliver the other side first, then explain the reason and the application entry point.
Usage guidelines
- Do not
list"force" keyword conditions onto the:listcommands; commands of this type only support time windows and pagination and cannot filter by keywords such as topic, creator, or participants; when encountering such needsmust switch tosearch。 - Do not treat keywords as time: when the user enters something like "last week's meeting with Zhang San",compound condition, identify "Zhang San" as a keyword and "last week" as time, and uniformly use
search。 - clarify first when ambiguous: If the user's wording is neither like a time nor like an explicit keyword (for example, only a number is given), first confirm whether it is a meeting number, meeting ID, or something else, then choose the corresponding command and parameters,do not infer arbitrarily。
Security Rules
-
It is forbidden to output AccessToken / RefreshToken to the terminal in plaintext.
-
It is strictly forbidden to expose to users
meeting_id, andmeeting_code(meeting number):meeting_idis an internal identifier used only for command-line parameter passing, and belongs toprivate fields; when displaying, repeating, or summarizing meeting information to users,uniformly usemeeting_code(meeting number), andmeeting_idmust not appear in replies (for example, all user-facing output scenarios such as response templates, secondary confirmation displays, and error feedback must comply). -
The following command operations must be confirmed twice: The following commands have irreversible effects on data or cause disturbance/notification to real people,before calling the command, the details of the operation to be performed must first be shown to the user, and it can be executed only after obtaining the user's explicit confirmation, and the confirmation step must not be skipped:
Command Risk Description meeting cancelCancel the meeting; irreversible meeting updateModify meeting information (time, topic, etc.), affecting all participants meeting invitees-addAdd invited members to the meeting; invitees will receive a meeting notification; before execution, the target meeting and member list must be displayed and explicit confirmation obtained meeting invitees-removeRemove invited members from the meeting meeting invitees-replaceReplace the meeting's invited member list as a whole (members not in the new list will be removed) control callActively call members to join the meeting, which initiates a meeting invitation call to the target member and causes actual disturbance to them control kickKick members out of the meeting, effective immediately;The target member's open_id/ms_open_idmust come fromreport participants, and it is strictly forbidden to usecontact searchresultsauth logoutClear local login credentials record permission-apply-commitFormally submit a recording permission application, which will trigger an approval process (you must first execute record permission-apply-prepareand show the application information to the user for confirmation)Confirmation Process:
- Show the user the operation to be performed and key information (use
meeting_codemeeting number to identify the meeting, and do not displaymeeting_id); when members are involved, the member echo format follows the "Member Echo Format" in the "Response Handling Rules"; - After displaying the informationmust end this round of reply, and wait for the user to explicitly reply with affirmative instructions such as "confirm", "yes", "yes", etc.; do not continue executing write operations in the same turn;
- execute the command only after receiving confirmation;
- If the user does not explicitly confirm or indicates cancellation, terminate the operation.
"Waiting" is a hard requirement — the following three practices are all violations, equivalent to skipping confirmation:
Violation Practice Manifestation Self-questioning and self-answering In the same reply, both asking "Confirm?" and then answering it yourself with "— Agree, submit" and then calling the command Fictitious user instruction Claims "based on your explicit instruction..." when that instruction does not exist in the conversation history Default auto-selection After listing the candidates, autonomously "defaults to option N" and continues execution Confirmation must come fromthe user's next real input, and must not be generated, inferred, or filled in on the user's behalf by the model itself.
- Show the user the operation to be performed and key information (use
-
When required parameters are missing, you must ask the user to supply them; filling them in on your own is prohibited: If the required parameters needed to execute a command are not provided by the user,you must not infer or fill in default values on your own, and you must clearly tell the user which parameters are missing and request that they be supplied; execute the command only after the user provides them.
-
Contacts search is limited to specific scenarios:
contact_search/contact_lookup_by_phone/contact_lookup_by_emailIt may be used only in the two scenarios of "meeting invitation" (e.g.,meeting invitees-add、meeting invitees-replace) and "calling a member into the meeting" (control call), to resolve a username to the correspondingopenId。It is strictly prohibited to callcontact searchin any other scenario (for example: merely to view someone's department/position, look up contact information, satisfy curiosity about someone's information, or other scenarios unrelated to meeting invitation/calling), and the contacts list must not be used as a general-purpose personnel information query interface. -
Hard constraint on the member source for removing a person during a meeting (
control kick):control kickThe--users/--sip-users/--pstn-usersparameter value ofopen_id/ms_open_id)must be obtained from the in-meeting participant list returned bytmeet report participantsRetrieved from the returned in-meeting attendee list,It is strictly prohibited to usecontact search/contact lookup-by-phone/contact lookup-by-emailand other contacts query results as the source for removing a person. Reason: the contacts list returns the organization's member directory, which does not mean they have joined the current meeting; moreover, removing a person requires distinguishing among the three identity types of regular member / Sip / Pstn, and this information can be accurately provided only byreport participants. Correct call order:tmeet report participants→ filter out the target participant by name or other description → confirm with the user →tmeet control kick。 -
Multiple results must be confirmed by the user; guessing on your own is prohibited: When any query/search-type command returns multiple candidate results(typically when
contact searchmatches multiple members with the same name/same department),it is strictly prohibited forthe model to autonomously select one entry based on position, department, hire date, match score, or any other dimension and continue with subsequent operations (such asmeeting invitees-add、control call、control kick, etc.). You must present the key information of the candidate entries to the user in a clear list and explicitly ask "Please confirm which one you want to select"; continue execution only after the user has explicitly specified one. Even if one of the results appears to be "obviously a better match", you must still wait for the user's confirmation and must not skip this step.
Parameter specifications
The following parameter rules are common to all commands, including the time parameter format, output control parameters (--format / --compact), and pagination parameters.
Time format
All time parameters use the ISO 8601 format, and the following two are supported:
| Format | Example |
|---|---|
| With time zone (with seconds) | 2026-03-12T14:00:00+08:00 |
| With time zone (without seconds) | 2026-03-12T14:00+08:00 |
Note: A date-only format (such as
2026-03-12) is not supported; time and time zone information must be included.
Time logic validation: If the end time provided by the user ≤ the start time (e.g., "from 4 o'clock to 3 o'clock"),do not infer the user's intent on your own, you must first confirm with the user whether it spans across days or is a typo, and then execute the command.
--format: Output JSON form
Used to control the layout form of the output JSON,does not change the field content. The output structure is uniformly {trace_id, message, data}。
| Value | Meaning | Applicable scenario |
|---|---|---|
json(default) |
Single-line compact JSON, small size, easy for pipeline passing | Model parsing, script processing,jq filtering |
json-pretty |
Multi-line indented JSON, highly readable | When the raw result needs to be presented directly to the user for reading |
Exception:
eventsubcommand family (event list/event schema/event consume/event status/event stop) outputs bare JSON,without the{trace_id, message, data}envelope—for example,event consumeeach line of NDJSON is in the form{event, trace_id, payload},event listis directly[{...}, ...]. For the specific form, see references/tmeet-event.md。
Usage example:
# 默认紧凑格式(模型解析场景推荐,省略 --format 即可)
tmeet meeting get --meeting-id 123456789
# 美化缩进格式(需要直接展示给用户阅读时使用)
tmeet meeting list --start 2026-03-12T00:00:00+08:00 --end 2026-03-12T23:59:59+08:00 --format json-pretty
Usage guidelines:
- When the model parses tool outputprefer to use the default
json, no need to explicitly pass--format;- Only when the user explicitly requests "display as beautified/formatted JSON" or the raw JSON needs to be presented to the user in full, append
--format json-pretty;- Even when using
json-pretty, the response handling rules still apply—only show key information, do not arbitrarily aggregate or sort。
--compact: Streamlined response fields
Boolean switch (default false), used totrim fields in the response body data fields in, keeping only a small number of fields necessary for the command's business, thereby significantly reducing the output token count.
- After enabling, the middleware will, based on the current command's API annotation, pull the "compact fields list" (compact fields) from the remote end, and perform field retention on the response
dataaccording to that list;trace_id、messagetop-level fields such as these are not affected. - If the current command has not declared an API annotation, or the remote pull fails, the middleware willtransparently pass through, will not block the main flow, and in this case the output is equivalent to not enabling
--compact. - With
--formatmutually independent:--formatdetermines the JSON layout,--compactdetermines the number of returned fields, and the two can be used simultaneously.
Usage example:
# 仅返回必要字段(推荐模型解析场景使用,节省 token)
tmeet meeting list --start 2026-03-12T00:00:00+08:00 --end 2026-03-12T23:59:59+08:00 --compact
# 同时启用精简字段 + 美化排版(便于用户直接阅读关键信息)
tmeet record list --meeting-id 123456789 --compact --format json-pretty
Usage guidelines:
- Query-type commands should be enabled first: When the model calls query/read-type commands,append by default
--compactto reduce context usage;- When not to use: When the user explicitly requests "complete results", "raw fields", or needs a certain non-essential field,do notuse
--compact。eventsubcommand family not applicable:event list/event schema/event consume/event status/event stopoutputdoes not go through the compact middleware, adding--compactwill not have any streamlining effect. To streamline event output, please switch toevent consume --jqfor projection.
Pagination
All query/list commands that support pagination uniformly adopt the --page-token + --page-size scheme.
Note:
record transcript-getthe command's--pid(start paragraph ID) and--limit(number of paragraphs to query)are not part of thegeneral pagination parameters; they are independent parameters used by that command for paragraph positioning,have not been deprecated, and can be used normally。
The original --page / --pos / --size parameters have all been marked asdeprecated, and are retained only for compatibility. Even if the user's dialogue uses expressions such as "page X" or "offset Y items", it should be implemented using the --page-token pagination strategy, andmust notuse deprecated parameters.
| Parameter | Description |
|---|---|
--page-token <token> |
Pagination cursor.Do not pass it on the first query; when paging, pass in the value of data.next_page_token from the previous response exactly as is |
--page-size <n> |
Number per page. Default values and upper limits differ by command; see each subcommand's documentation for details |
Usage Guidelines:
- Prefer using
--page-tokenfor paging: when calling the next page, you must take the value from thedata.next_page_tokenfield of the previous response and pass it to--page-token, and must not concatenate, increment, or guess this value yourself. - Determining Reaching the Last Page: when the
next_page_tokenin the response is an empty string or the field is missing, this is the last page, and paging should not continue. - When there is too much data, you must first ask the user whether to continue paging; unauthorized continuous paging is prohibited: if in the results returned by the first query
next_page_tokenis not empty (that is, there is still subsequent data), and the user's original request does not explicitly require exhaustive expressions such as "all / every / complete",you must not continuously call the next page on your own; you must first show the user the key information of the current page and a prompt that "there are more results", and ask the user whether to continue paging (for example, "The first N results have been displayed so far, and there is more data. Do you want to view the next page?"). After receiving an explicit affirmative instruction from the user (such as "continue", "yes", "next page", "all", etc.), then use thenext_page_tokenfrom the previous page to initiate the next query; if the user indicates stop or does not explicitly confirm, then terminate paging. When the user has explicitly requested "all / every", you may continuously page untilnext_page_tokenis empty; however,when more than 5 pages have been continuously paged or the cumulative number of items exceeds 200, you must proactively prompt and ask the user whether to continue. record transcript-searchdoes not currently support pagination, and no pagination parameters need to be passed.
Typical Paging Flow:
# 1) 首次查询(不传 --page-token)
tmeet record list --meeting-id "100000000" --page-size 30 --compact
# 2) 从响应中取出 data.next_page_token,继续翻页
tmeet record list \
--meeting-id "100000000" \
--page-token "<next_page_token>" \
--page-size 30 --compact
Response Handling Rules
-
Show only key information: unless the user explicitly requests otherwise, show only the core fields directly related to the user's question, and do not output redundant fields.
-
Do not aggregate or sort without authorization: without the user's request, do not perform any aggregation statistics or sorting operations on the returned results; present the original results faithfully.
-
Member echo format (hard constraint for outputs involving members): When the output object contains "people" (including but not limited to all user-facing member output scenarios such as secondary confirmation displays, query result displays, and member list echo after a successful operation), each membermustbe echoed in the
姓名(<标识>)format:- Nameis taken according to the following priority: ① the display name field in the contacts / participant list response; ② if the response does not return a name field (typically such as
contact searchunique match returning onlyopen_id、contact lookup-by-phone/emailreturning onlyopen_id),use the search keyword entered by the user this time(such as the value of--username) as the name; ③ if neither can be obtained, mark it as未知成员。strictly prohibitedto useopen_id/ms_open_id/userid/ email prefix / nickname in place of the name; - Identifier in parenthesesfrom
部门、职位、open_idmiddlechoose any one itemto fill in (fill in if available; omitting the parentheses is allowed only when none are available),completeness is not required, priority:部门>职位>open_id; - If a member's name is missing (following branch ③ above), mark it as
未知成员(<open_id>),must notomit the identifier; - Example:
张三(技术工程事业群/腾讯会议)、李四(高级工程师)、张三(ou_abc123...)。
- Nameis taken according to the following priority: ① the display name field in the contacts / participant list response; ② if the response does not return a name field (typically such as
-
Secondary confirmation template for invitee management write operations (strong constraint): Before executing
meeting invitees-add/meeting invitees-remove/meeting invitees-replacethe above three commands,mustdisplay to the user according to the following template, none may be omitted; the members in the template strictly follow the above "member echo format" (姓名(<标识>)), meetings always usemeeting_code(meeting number),strictly prohibitedfrom appearingmeeting_id。- Operation: Will invite <members to be invited, strictly following the above "member echo format", separate multiple people with the Chinese enumeration comma "、"> <action description> "<meeting topic>" (meeting number <meeting_code>) <object description>
- Impact: <select the corresponding impact description according to the command>
-
Reply template for meeting member change operations (strong constraint): After successful execution of
meeting invitees-add/meeting invitees-remove/meeting invitees-replace, the replymustorganize the fields strictly according to the following template, and display only these fields:- Meeting topic
- Meeting time(start time ~ end time, including time zone)
- Meeting number(
meeting_code,strictly prohibitedfrom displayingmeeting_id) - Join link(
join_url) - Invited members(after the operation is completed,the current completelist of invited members)
Display rules (strictly follow):
- Each member in "Invited members" strictly follows the above "member echo format"(
姓名(<标识>)), if you do not have the mapping foropen_id → 姓名at hand,first callmeeting invitees-listto obtain the latest invited listand then echo it; - If the response does not directly return basic fields such as the meeting topic, meeting number, and join link, they should be completed through
meeting get --meeting-id, and no field may be omitted or replaced with-/N/Aplaceholder.
Automatic feedback rules
When the Agent encounters any of the following situations while using the tmeet CLI, it should recognize them and prepare to call tmeet tshoot feedback to report the issue to the platform,but before actually performing the report, it must first confirm with the user a second time, and only execute after obtaining explicit consent。
Command parameters,
--categorythe complete trigger scenarios, character length constraints, and examples of the enum: seereferences/tmeet-tshoot.md。
Trigger timing (5 categories)
The Agent should recognize one of the following 5 scenarios and trigger feedback:tool_not_found(tool missing) / tool_error(tool error) / tool_inadequate(insufficient capability) / unexpected_result(abnormal result) / suggestion(improvement suggestion). For the detailed trigger scenarios and judgment criteria of each category, see the table references/tmeet-tshoot.md «--category enum values».
Invocation guidelines (hard constraints at the decision layer)
- Must confirm a second time before reporting: after recognizing the above trigger conditions,first show the user the content to be fed back(including
--category、--intent、--actions-tried、--resultand other key fields), and explicitly ask the user whether they agree to report;only after receiving the user's explicit confirmation (such as "confirm", "yes", "yes", etc.)executetmeet tshoot feedback; if the user refuses or does not explicitly agree, do not report. After the report is completed,briefly inform the user"The issue has been reported to the platform for you." - Does not replace normal error handling: feedback is only used to inform the platform,and must not be used to bypass the user's original task. If there are still executable alternatives (such as switching to another command or retrying with additional parameters), you shouldtry to resolve it first, and if it cannot be resolved, then ask the user whether to report.
- Fill in the context truthfully:
--intentmust truthfully state the user's original intent;--actions-triedstate the commands already attempted;--resultstate the blocking point or error message; when specific commands are involved, fill in--tool-name; when there is an error code, fill in--error-code。It is strictly forbidden to fabricate or fill in irrelevant content。 - Strong privacy desensitization constraints: in the feedback content,it is strictly forbidden to disclose the user's name / phone number / meeting number / meeting link / meeting subject / participantsand other information involving the user's personal privacy. If relevant content must be cited to help explain the issue,it must first be desensitized by masking with asterisks, encryption, etc.(for example: name
张三→张*, mobile number13800138000→138****8000, meeting number123456789→12****789, meeting subjectQ2 项目复盘会→Q* 项目***会) before writing it into--intent/--actions-tried/--resultand other fields. - Login prerequisite: this command depends on the login state. If the user is not logged in yet, first guide them to execute
tmeet auth login, and after successful login, initiate the feedback. - Deduplication and moderation: in the same user session, forthe same issuereport onlyonce, to avoid repeated spamming; report different issues separately and independently.
Common errors
| Error symptom | Cause | Solution |
|---|---|---|
user config is empty |
Not logged in | Execute tmeet auth login |
--start format error |
Invalid time format (e.g., missing time zone) | Use 2026-03-12T14:00:00+08:00 format instead |
user has been initialized |
Already logged in, executing login again | Use directly, or logout first then login |
Error code 500284 / 该功能暂不可使用 |
The current user does not have permission to use this feature (possibly not in the gray release, insufficient commercial rights, etc.) | Retrying is strictly prohibited(retrying is meaningless; permission issues are not temporary failures); inform the user truthfully: "This feature is currently unavailable for this account, possibly because the gray release has not been enabled or commercial rights are insufficient; the right of final interpretation belongs to Tencent Meeting." Do not speculate on the specific reason or promise any method of enabling it. |
event consume related errors (ready flag does not appear / exit code 1 / event stop returns refused , etc.) |
— | See references/tmeet-event.md "Typical Fault Quick Reference" |