dsh-plugin-feishu
Feishu (Lark) tools for DeepSeek Harness: 24 feishu_* tools covering chat
messages, cloud documents (including images, generated charts and tables),
bitable records, spreadsheet ranges and drive permissions — through your own
enterprise self-built Feishu app. Pure Node, no external dependencies, no Python.
Install
Requires DeepSeek Harness with plugin support (dsh plugin).
dsh plugin --profile web add dsh-plugin-feishu # once published to npm
dsh plugin --profile web add github:2286893544/feishu-dsh-plugin # straight from GitHub
Reload the profile, then configure the plugin on its own settings page:
Settings → 飞书（Feishu）. The client half (client/client.js) contributes
that sidebar page (a settings.section, the same slot community plugins such as
the side card use); it writes into the feishu-bridge settings namespace, so a
saved change applies immediately:
| Field | Meaning |
|---|
appId | Feishu app ID of your self-built app (cli_...) |
appSecret | Feishu app secret — declared role("secret"), so it is redacted from every page read and shown as a password control that starts blank |
tenantDomain | Optional (acme.feishu.cn); auto-detected from the tenant API when empty |
defaultChatId | Optional (oc_...) used by the chat tools when a call omits chat_id |
On a host without a settings service the card is simply absent and the composed
configuration stands. Both fallbacks remain: the profile patch entry
(config: { appId, appSecret, ... }) and the environment variables
FEISHU_APP_ID, FEISHU_APP_SECRET, FEISHU_TENANT_DOMAIN.
The page also carries a 测试连接 (test connection) button. It POSTs to the
host route /dsh-plugin-feishu/test-connection, which exercises the stored
credentials against the Feishu API and answers with non-secret facts only: the
app name, the tenant name and domain, a masked key fingerprint
(pkQi…E2h (len 32)) and which configuration layers supplied the values. The raw
secret never leaves the host, and the route refuses GETs and cross-origin calls.
Tools (24)
Chat & messages — feishu_list_chats, feishu_list_chat_members,
feishu_send_text, feishu_send_post_message, feishu_read_chat_history,
feishu_recall_message
Documents (docx) — feishu_create_document, feishu_read_document,
feishu_read_document_blocks, feishu_append_document_blocks,
feishu_update_document_block, feishu_delete_document_block,
feishu_insert_table_into_document, feishu_insert_image_into_document,
feishu_insert_chart_into_document
Bitable — feishu_create_bitable, feishu_list_bitable_tables,
feishu_list_bitable_fields, feishu_write_bitable_record,
feishu_read_bitable_records
Sheets (classic v2 range API) — feishu_read_sheet_range,
feishu_write_sheet_range
Permissions — feishu_grant_document_access
A bundled feishu skill documents the workflows and known limits for the agent.
Notes and limits
- Charts are rendered locally (pure Node PNG encoder + 5×7 ASCII font); chart
titles and category labels are ASCII-only — non-ASCII characters are dropped.
For Chinese labels, generate the image yourself and use
feishu_insert_image_into_document.
- Spreadsheet tools use the classic sheets v2 range API.
- Only chats the bot belongs to can be read; other users' private chats are not
accessible through the Feishu API.
Prerequisites (Feishu side)
- Enterprise self-built app with the bot enabled.
- Published scopes for reading chat history (
im:message.group_msg …), docx
read/write, bitable, sheets and drive permission management; the bot must be a
member of the target chats, and documents the app must read must be shared
with it.
- Credentials stay local; never share
appSecret in chat or documents.
Development
node scripts/verify-registration.mjs # offline: registers 24 tools + skill, renders a chart
node scripts/smoke.mjs # live API check (needs credentials; optional doc id)
License
MIT
