Jev Gallery
中文EN

JEV GUIDE

Jev for Feishu message triage: personal tasks and urgency

A deadline in a group chat is not automatically your urgent task. Work with Jev evaluates selected messages alongside your role, then local code assigns a column. This guide separates triage from actually carrying out work.

Jev Gallery · Reviewed

Start with the person reading the message

“Confirm the proposal today” means something different to the owner, an observer or a former team member. Define a name, current responsibilities, relevant projects and the reference time. Supply enough context to identify ownership without uploading a full personal profile.

From four signals to four columns

The reviewed source asks about personal relevance, required action, urgency and reference value separately. Code uses the returned values to assign urgent, to-do, useful or skip, with some intermediate judgments flagged for review.

A message can be both informative and actionable. Separate signals allow the application to express a priority policy instead of assuming four mutually exclusive facts. Calibrate the defaults on your own labeled messages.

Source: Work with Jev — core.mjs ↗

Read from Feishu, then approve classification

The project reads a selected chat through Feishu CLI user authorization. Fetching messages and sending them to TypeSafe are separate actions; classification follows a user review of the proposed input. This does not grant access to chats outside the user’s permissions.

Check the local Node.js setup, CLI login and required read permissions. Start with the preset demo, then fetch a small real batch. The reviewed version handles text and rich text, not the contents of images, attachments or voice messages.

Source: Work with Jev — README.md ↗

Test urgency with counterexamples

These fixtures are editorial designs, not live Jev responses. Expected labels depend on the actual role and surrounding messages.

Test urgency with counterexamples
Synthetic message and readerHuman expectationFailure to investigate
Shoot owner: “We are waiting for your camera-position approval.”UrgentMissing a blocker
Observer: “@Lin, send your design today.”Not a personal taskAssigning someone else’s work
Owner: “Prepare the minutes by next Friday.”Task, not automatically urgentDeadline mistaken for urgency
“The proposal request has been canceled.”Review the existing taskCreating a new task from cancellation

Keep task state and context boundaries explicit

Users can correct labels and add messages to a cross-chat queue without asking the model again. Completion and ordering remain user-controlled. A refresh is not an instruction to reassign ownership or reset task state.

Messages are judged in batches without shared context between batches. A cancellation may be separated from its original task. Do not promise automatic reconciliation of historical work; investigate errors before changing the message window, role description or rules.

Source: Work with Jev — README.md ↗

Common questions and rollout checks

Does it support WeCom or Slack? Those integrations are not implemented in the reviewed version. An adapter interface is not a working platform integration.

Does it reply or execute tasks? No. It sorts messages and supports manual task management. Label a permitted sample by role, measure missed tasks and false alarms, and verify the review path before expanding use.

We reviewed pinned documentation and source, not live Feishu authorization, model accuracy or end-to-end latency.

Source: Work with Jev — README.md ↗

Explore these projects next

Continue reading