Sibling skills (local only)
Sibling CloudBase skills ship beside this skill. Use local relative paths such as ../auth-tool-cloudbase/SKILL.md.
If a referenced sibling skill file is missing from this environment, ask the user to install the full CloudBase plugin (or the missing skill). Do not HTTP-fetch remote skill or protocol markdown into the agent context.
CloudBase Document Database Web SDK
Activation Contract
Use this first when
- A browser or Web app must read or write CloudBase document database data through
@cloudbase/js-sdk.
- The request mentions
app.database(), db.collection(), .where(), .watch(), pagination, aggregation, or geolocation queries in a Web frontend.
Read before writing code if
- The task is clearly browser-side, but you still need to decide between Web SDK, Mini Program SDK, or backend access.
- The request touches login state, collection permissions, or realtime updates.
Then also read
- Web login and caller identity ->
../auth-web-cloudbase/SKILL.md
- General Web app structure ->
../web-development/SKILL.md
- Mini Program database code ->
../cloudbase-document-database-in-wechat-miniprogram/SKILL.md
Do NOT use for
- Mini Program code using
wx.cloud.database().
- Server-side or cloud-function database access.
- SQL / MySQL database operations.
- Pure resource-permission administration with no browser SDK code.
- NEW business tables that the task explicitly asks to put in CloudBase PostgreSQL (CloudBase PG). Before applying this skill, call
queryEnv(action="info", envId=...) and read EnvInfo.RuntimeBackends. If postgresql === true AND the task asks for a new business table to live in PG, switch to the postgresql-development-cloudbase skill for that table: it goes through app.rdb(), uses PG row-level security (CREATE POLICY), and uploads via app.storage.from('<bucket>').upload('<key>', file) against an explicitly-created pgstore bucket.
- Existing NoSQL collections in the same env keep using THIS skill — PG and NoSQL coexist in CloudBase PG environments. The rule is "follow the task / existing surface", not "PG env forbids NoSQL".
- Moving existing document data onto PG without changing the client code is pgdoc, not a rewrite to
app.rdb(). Read ../postgresql-development-cloudbase/references/pgdoc.md: SDK calls and HTTP paths stay the same, but CUSTOM security rules no longer apply and data is not migrated automatically.
SDK Code vs MCP Tools
When to write SDK code (use this skill):
- The task explicitly asks to "modify code" or "use SDK"
- The task asks to implement app/frontend logic
- The task mentions specific SDK methods like
db.collection().add(), .get(), .update()
- The context shows an existing Web project with SDK initialization (e.g.,
index.js already has cloudbase.init())
When to use MCP tools instead:
- The task asks to manage CloudBase resources (create collection, set permissions, etc.)
- The task involves admin/management operations without writing app code
- The task mentions tools like
writeNoSqlDatabaseContent, managePermissions, etc.
Key distinction: If the user says "使用 JS SDK 执行 XX 操作" (use JS SDK to perform XX operation) or "修改代码" (modify code), write SDK code in the project files. Do not use MCP database write tools for app-level data operations.
Common mistakes / gotchas
- Querying before the user is signed in when the collection rules require identity.
- Using
wx.cloud.database() or Node SDK patterns in browser code.
- Initializing CloudBase lazily with dynamic imports instead of a shared synchronous app instance.
- Treating security rules as result filters rather than request validators.
- Expecting a
CUSTOM security rule to take effect immediately after you call managePermissions(updateResourcePermission). The backend caches rule evaluators for 2–5 minutes; first writes after a rule change may silently fail or be rejected with DATABASE_PERMISSION_DENIED even when the expression is correct. Either (a) wait a few minutes and retry the same write before assuming the rule is wrong, or (b) verify the rule is live by reading result.code / result.message on every write and by doing a get() round-trip on the just-written _id; do not treat a resolved promise as success. See security-rules.md → "Propagation And Verification" for the full pattern.
- Misreading the return shape of
db.collection(...).add(...). In the CloudBase Web SDK, the created document ID is exposed at top-level result._id, not result.id, result.data.id, or result.insertedId.
- For CMS-style collections that need app-level admin users to edit/delete all records while editors can only edit/delete their own records, do not oversimplify the rule to
READONLY. A validated pattern is a CUSTOM rule that reads role from user_roles by auth.uid and combines it with doc.authorId == auth.uid, while frontend writes can stay on .doc(id).update() / .doc(id).remove().
- Forgetting pagination or indexes for larger collections.
Minimal checklist
- Confirm this is browser-side document database work.
- Initialize CloudBase once and reuse the same
app / db instance.
- Verify auth expectations before CRUD.
- Read the right companion reference file for the specific operation.
Overview
This skill covers browser-side document database usage via @cloudbase/js-sdk.
Use it for:
- CRUD in a Web app
- complex queries and pagination
- aggregation
- realtime listeners with
watch()
- geolocation queries
Canonical initialization
import cloudbase from "@cloudbase/js-sdk";
const app = cloudbase.init({
env: "your-env-id"
});
const db = app.database();
const _ = db.command;
Important rules:
- Sign in before querying if the collection rules require identity.
- Keep a single shared app/database instance.
- Do not hide initialization inside ad-hoc async loaders unless the framework truly requires it.
Quick routing
- CRUD ->
./crud-operations.md
- Complex queries ->
./complex-queries.md
- Pagination ->
./pagination.md
- Aggregation ->
./aggregation.md
- Realtime listeners ->
./realtime.md
- Geolocation ->
./geolocation.md
- Security rules ->
./security-rules.md
Working rules for a coding agent
Start from the auth model
- If the page relies on logged-in user identity, read the Web auth skill before writing database code.
Keep browser code browser-native
- Use
app.database() and collection references.
- Do not mix in MCP management flows or SQL mental models.
Respect security rules
- Collection rules can reject requests before data is read.
- If the requirement is simple owner-only write access,
READONLY can be enough.
- If the requirement is “app-level admin can edit/delete all, editor only own”, use a
CUSTOM rule. A validated CMS pattern is get(`database.user_roles.${auth.uid}`).role == 'admin' || doc.authorId == auth.uid.
- For that CMS pattern, frontend writes can stay on
.doc(id).update() / .doc(id).remove().
- Reuse whichever role collection already exists and can be addressed by
_id == auth.uid. In this CMS pattern, user_roles keyed by uid is acceptable.
- A
get() whose target document is missing fails the request with HTTP 500 — it does not evaluate to false, and || does not skip it. If the app cannot guarantee a role document per user, keep the role list in one shared document and test auth.uid in get(`database.admin_registry.lock`).admins instead.
- If the task fails with permission issues, inspect the rule model rather than assuming the query syntax is wrong.
Return user-friendly errors
- Database errors must become readable UI or application errors, not silent failures.
- For writes, do not treat a resolved promise as success by default. Check write result fields such as
updated / deleted or surfaced code / message.
Persist IDs from create operations correctly
- For Web SDK
.add(...), the newly created document ID is result._id.
- Do not look for the ID under
result.id, result.data, or other driver-specific fields.
Quick examples
Simple query
const result = await db.collection("todos")
.where({ completed: false })
.get();
Create and capture document ID
const result = await db.collection("posts").add({
title: "New article",
content: "...",
createdAt: new Date()
});
const articleId = result._id;
Ordered pagination
const result = await db.collection("posts")
.orderBy("createdAt", "desc")
.skip(20)
.limit(10)
.get();
Field selection
const result = await db.collection("users")
.field({ name: true, email: true, _id: false })
.get();
Best practices
- Define collection-level types or model wrappers in the app code.
- Use meaningful collection naming conventions.
- Select only required fields.
- Add indexes for frequent filters or sort keys.
- Pair frontend CRUD with explicit permission design.
- Use pagination instead of unbounded reads.
Error handling
try {
const result = await db.collection("todos").get();
console.log(result.data);
} catch (error) {
console.error("Database error:", error);
}
When the SDK returns an operation result, check error indicators and translate them into readable application behavior.
1---2name: cloudbase-document-database-web-sdk-63description: Use CloudBase document database Web SDK only for confirmed NoSQL collection work. Query, create, update, and delete document data; if the task mentions PostgreSQL / CloudBase PG / app.rdb(), route to postgresql-development instead.4version: 2.37.65license: MIT6alwaysApply: false7metadata:8 source: https://skillmd.com/skills/tencentcloudbase/cloudbase-document-database-web-sdk-69---1011## Sibling skills (local only)1213Sibling CloudBase skills ship beside this skill. Use local relative paths such as `../auth-tool-cloudbase/SKILL.md`.1415If a referenced sibling skill file is missing from this environment, ask the user to install the full CloudBase plugin (or the missing skill). Do **not** HTTP-fetch remote skill or protocol markdown into the agent context.1617# CloudBase Document Database Web SDK1819## Activation Contract2021### Use this first when2223- A browser or Web app must read or write CloudBase document database data through `@cloudbase/js-sdk`.24- The request mentions `app.database()`, `db.collection()`, `.where()`, `.watch()`, pagination, aggregation, or geolocation queries in a Web frontend.2526### Read before writing code if2728- The task is clearly browser-side, but you still need to decide between Web SDK, Mini Program SDK, or backend access.29- The request touches login state, collection permissions, or realtime updates.3031### Then also read3233- Web login and caller identity -> `../auth-web-cloudbase/SKILL.md`34- General Web app structure -> `../web-development/SKILL.md`35- Mini Program database code -> `../cloudbase-document-database-in-wechat-miniprogram/SKILL.md`3637### Do NOT use for3839- Mini Program code using `wx.cloud.database()`.40- Server-side or cloud-function database access.41- SQL / MySQL database operations.42- Pure resource-permission administration with no browser SDK code.43- **NEW business tables that the task explicitly asks to put in CloudBase PostgreSQL (CloudBase PG).** Before applying this skill, call `queryEnv(action="info", envId=...)` and read `EnvInfo.RuntimeBackends`. If `postgresql === true` AND the task asks for a new business table to live in PG, switch to the `postgresql-development-cloudbase` skill for that table: it goes through `app.rdb()`, uses PG row-level security (`CREATE POLICY`), and uploads via `app.storage.from('<bucket>').upload('<key>', file)` against an explicitly-created pgstore bucket.44 - Existing NoSQL collections in the same env keep using THIS skill — PG and NoSQL coexist in CloudBase PG environments. The rule is "follow the task / existing surface", not "PG env forbids NoSQL".45- **Moving existing document data onto PG without changing the client code** is pgdoc, not a rewrite to `app.rdb()`. Read `../postgresql-development-cloudbase/references/pgdoc.md`: SDK calls and HTTP paths stay the same, but CUSTOM security rules no longer apply and data is not migrated automatically.4647### SDK Code vs MCP Tools4849**When to write SDK code (use this skill):**50- The task explicitly asks to "modify code" or "use SDK"51- The task asks to implement app/frontend logic52- The task mentions specific SDK methods like `db.collection().add()`, `.get()`, `.update()`53- The context shows an existing Web project with SDK initialization (e.g., `index.js` already has `cloudbase.init()`)5455**When to use MCP tools instead:**56- The task asks to manage CloudBase resources (create collection, set permissions, etc.)57- The task involves admin/management operations without writing app code58- The task mentions tools like `writeNoSqlDatabaseContent`, `managePermissions`, etc.5960**Key distinction:** If the user says "使用 JS SDK 执行 XX 操作" (use JS SDK to perform XX operation) or "修改代码" (modify code), write SDK code in the project files. Do not use MCP database write tools for app-level data operations.6162### Common mistakes / gotchas6364- Querying before the user is signed in when the collection rules require identity.65- Using `wx.cloud.database()` or Node SDK patterns in browser code.66- Initializing CloudBase lazily with dynamic imports instead of a shared synchronous app instance.67- Treating security rules as result filters rather than request validators.68- **Expecting a `CUSTOM` security rule to take effect immediately after you call `managePermissions(updateResourcePermission)`.** The backend caches rule evaluators for **2–5 minutes**; first writes after a rule change may silently fail or be rejected with `DATABASE_PERMISSION_DENIED` even when the expression is correct. Either (a) wait a few minutes and retry the same write before assuming the rule is wrong, or (b) verify the rule is live by reading `result.code` / `result.message` on every write and by doing a `get()` round-trip on the just-written `_id`; do not treat a resolved promise as success. See `security-rules.md` → "Propagation And Verification" for the full pattern.69- Misreading the return shape of `db.collection(...).add(...)`. In the CloudBase Web SDK, the created document ID is exposed at top-level `result._id`, not `result.id`, `result.data.id`, or `result.insertedId`.70- For CMS-style collections that need **app-level admin users** to edit/delete all records while editors can only edit/delete their own records, do not oversimplify the rule to `READONLY`. A validated pattern is a `CUSTOM` rule that reads role from `user_roles` by `auth.uid` and combines it with `doc.authorId == auth.uid`, while frontend writes can stay on `.doc(id).update()` / `.doc(id).remove()`.71- Forgetting pagination or indexes for larger collections.7273### Minimal checklist7475- Confirm this is browser-side document database work.76- Initialize CloudBase once and reuse the same `app` / `db` instance.77- Verify auth expectations before CRUD.78- Read the right companion reference file for the specific operation.7980## Overview8182This skill covers **browser-side document database usage** via `@cloudbase/js-sdk`.8384Use it for:8586- CRUD in a Web app87- complex queries and pagination88- aggregation89- realtime listeners with `watch()`90- geolocation queries9192## Canonical initialization9394```javascript95import cloudbase from "@cloudbase/js-sdk";9697const app = cloudbase.init({98 env: "your-env-id"99});100101const db = app.database();102const _ = db.command;103```104105Important rules:106107- Sign in before querying if the collection rules require identity.108- Keep a single shared app/database instance.109- Do not hide initialization inside ad-hoc async loaders unless the framework truly requires it.110111## Quick routing112113- CRUD -> `./crud-operations.md`114- Complex queries -> `./complex-queries.md`115- Pagination -> `./pagination.md`116- Aggregation -> `./aggregation.md`117- Realtime listeners -> `./realtime.md`118- Geolocation -> `./geolocation.md`119- Security rules -> `./security-rules.md`120121## Working rules for a coding agent1221231. **Start from the auth model**124 - If the page relies on logged-in user identity, read the Web auth skill before writing database code.1251262. **Keep browser code browser-native**127 - Use `app.database()` and collection references.128 - Do not mix in MCP management flows or SQL mental models.1291303. **Respect security rules**131 - Collection rules can reject requests before data is read.132 - If the requirement is simple owner-only write access, `READONLY` can be enough.133 - If the requirement is “app-level admin can edit/delete all, editor only own”, use a `CUSTOM` rule. A validated CMS pattern is ``get(`database.user_roles.${auth.uid}`).role == 'admin' || doc.authorId == auth.uid``.134 - For that CMS pattern, frontend writes can stay on `.doc(id).update()` / `.doc(id).remove()`.135 - Reuse whichever role collection already exists and can be addressed by `_id == auth.uid`. In this CMS pattern, `user_roles` keyed by uid is acceptable.136 - A `get()` whose target document is missing fails the request with HTTP 500 — it does not evaluate to `false`, and `||` does not skip it. If the app cannot guarantee a role document per user, keep the role list in one shared document and test ``auth.uid in get(`database.admin_registry.lock`).admins`` instead.137 - If the task fails with permission issues, inspect the rule model rather than assuming the query syntax is wrong.1381394. **Return user-friendly errors**140 - Database errors must become readable UI or application errors, not silent failures.141 - For writes, do not treat a resolved promise as success by default. Check write result fields such as `updated` / `deleted` or surfaced `code` / `message`.1421435. **Persist IDs from create operations correctly**144 - For Web SDK `.add(...)`, the newly created document ID is `result._id`.145 - Do not look for the ID under `result.id`, `result.data`, or other driver-specific fields.146147## Quick examples148149### Simple query150151```javascript152const result = await db.collection("todos")153 .where({ completed: false })154 .get();155```156157### Create and capture document ID158159```javascript160const result = await db.collection("posts").add({161 title: "New article",162 content: "...",163 createdAt: new Date()164});165166const articleId = result._id;167```168169### Ordered pagination170171```javascript172const result = await db.collection("posts")173 .orderBy("createdAt", "desc")174 .skip(20)175 .limit(10)176 .get();177```178179### Field selection180181```javascript182const result = await db.collection("users")183 .field({ name: true, email: true, _id: false })184 .get();185```186187## Best practices1881891. Define collection-level types or model wrappers in the app code.1902. Use meaningful collection naming conventions.1913. Select only required fields.1924. Add indexes for frequent filters or sort keys.1935. Pair frontend CRUD with explicit permission design.1946. Use pagination instead of unbounded reads.195196## Error handling197198```javascript199try {200 const result = await db.collection("todos").get();201 console.log(result.data);202} catch (error) {203 console.error("Database error:", error);204}205```206207When the SDK returns an operation result, check error indicators and translate them into readable application behavior.