Skip to main content

NineData Skill

NineData-Skill gives AI Agents and vibe-coding tools governed access to NineData platform capabilities. It uses NineData authentication, permission checks, SQL rule checks, data masking, throttling, approval, and audit logging. Typical scenarios include listing data sources, executing read-only SQL, submitting SQL tasks, importing data, exporting query results, and submitting Schema Design and Release tasks.

What NineData Skill Can Do​

  • SQL execution: Execute explicit SQL in a permitted data source context, or retrieve execution plans and related information.
  • SQL task: Submit SQL to a secure approval workflow for scenarios that require rule checks, approval, and audit.
  • Data export: Export read-only query results to CSV, Excel, SQL, JSON, XML, or text files and track approval, execution, logs, and file status.
  • Data import: Use data-import to upload local SQL, CSV, Excel, JSON, JSONL, or NDJSON files and submit import tasks. Check file status and sensitive-data checks, and track approval, execution, details, and process logs.
  • Schema Design and Release: Submit Schema Design and Release tasks through NineData-Skill to advance database changes across multiple environments. See Schema Design and Release for the scope and workflow.
  • Platform capability calls: Let AI Agents call NineData platform capabilities through OpenAPI, such as listing data sources available to the current account, checking database status, or validating development data. See Retrieve Data Source List for the API reference.
  • Security controls: Skill calls are governed by NineData account identity, organization permissions, SQL rules, data masking, throttling, and audit logs.

Skill tools​

ToolPurpose
sql-executeExecute read-only DQL or metadata checks and return the platform decision and result preview.
sql-taskSubmit and operate SQL tasks that require rule checks, approval, or execution control.
data-exportSubmit and operate Data Export workflows for read-only query results.
data-importUpload local data files and submit import tasks; check file status, sensitive-data checks, approval and execution status, and process logs.

Use cases​

Use caseDescription
Discover accessible data sourcesList the data sources available to the current OpenAPI credential before selecting a target environment. You can also call the Retrieve Data Source List OpenAPI.
Run read-only queriesExecute explicit query SQL within the authorized scope and return a result preview.
Analyze execution plansRetrieve execution plans or diagnostic information for read-only SQL.
Submit SQL tasksRoute SQL that changes data, schema, permissions, or transaction state through a governed workflow.
Export query resultsCreate a Data Export workflow for read-only query results and follow platform approval, execution, and file permissions.
Import dataImport data files into a target data source under permission and approval controls, then track the task status.
Schema Design and ReleaseSubmit Schema Design and Release tasks in a multi-environment database development workflow. See Schema Design and Release for scope and steps.
Assist development workflowsValidate development data within NineData identity and permission boundaries.

Security considerations​

  • Confirm the target data source, complete SQL, operation reason, and account permissions before an Agent calls Skill.
  • Use SQL execution only for read-only DQL and metadata checks.
  • Use SQL task for SQL that can change data, schema, permissions, transaction state, or stored procedure state.
  • Use Data export when a read-only query must produce a file. File download and encryption-password retrieval require separate authorization.
  • For Data Import, upload the local file and confirm that it is available before submitting the import task. A successful upload does not mean the data has been imported. A user must explicitly confirm sensitive-data warnings; the Agent must not disable sensitive-data checks automatically.
  • Do not store database usernames or passwords in Skill or let the Agent connect directly to a database.
  • Do not expose accessKeySecret, cookies, tokens, signatures, or database credentials in chats, logs, screenshots, or commits.
  • Do not bypass a platform rejection, confirmation requirement, SQL task requirement, or approval requirement.
  • Approval actions such as approve, reject, and transfer-approval require explicit authorization for the exact action.
caution

Keep OpenAPI credentials in a restricted local configuration. Treat exported files and encryption passwords as sensitive data. Do not place them in documentation, screenshots, chats, or repositories.

Before you begin​

  • You are logged in to the NineData console.
  • Your account has permission to access the Skill page.
  • Create an AccessKey for the calling scenario in Account > AccessKey Management in the NineData console, and save the AccessKey and SecretKey from the creation result page. For details, see Manage AccessKey.
  • To execute SQL or submit SQL tasks through Skill, your account must have the required permissions on the target data source and DevOps workflow.

View Skill Capabilities​

  1. Sign in to the NineData Console.

  2. From the left sidebar, open Skill.

  3. On the NineData-Skill page, view the supported Skill capabilities.

    The page shows what NineData-Skill is used for and lists the feature name, description, and release time of each capability.

Page Actions​

On the NineData-Skill page toolbar, use these actions:

ActionDescription
DescriptionView the NineData-Skill version, supported capabilities, main return information, and usage boundaries.
DownloadDownload ninedata-skill.zip.
Quick setup guideView installation paths and validation prompts for Claude Code, Codex, Cursor, Hermes, OpenClaw, Trae, Qoder, and Open Code, then follow the guide for the selected client.

Download and Configure Skill​

  1. On the NineData-Skill page toolbar, click the download icon to download ninedata-skill.zip.

  2. Open the quick setup guide, select the target client, and follow that client's instructions to install NineData-Skill in the corresponding directory. Example installation paths:

    ClientInstallation path example
    Claude Code~/.claude/skills/ninedata-skill/
    Codex~/.codex/skills/ninedata-skill/
    Cursor~/.cursor/skills/ninedata-skill/
    Hermes Agent~/.hermes/skills/ninedata-skill/
    OpenClaw~/.openclaw/skills/ninedata-skill/
    Trae~/.trae/skills/ninedata-skill/
    Qoder~/.qoder/skills/ninedata-skill/
    Open Code~/.opencode/skills/ninedata-skill/
  3. Follow the selected client's instructions to configure NineData-Skill, set any required local command permissions, and restart the client when required. Keep OpenAPI credentials only in a restricted local configuration.

  4. Ask the Agent to use NineData Skill.

    Use ninedata-skill to list available MySQL data sources.

Validate Configuration​

After configuration, ask NineData-Skill to list the data sources accessible to the current account. The request is handled through the NineData platform and does not make the Agent connect directly to a database.

Query Data with Skill​

After Skill is configured, describe your query requirement in natural language in the AI Agent client. For example:

Use ninedata-skill to query the order count in the production MySQL data source.

The Agent should identify accessible data sources and present the full SQL that will be executed. Confirm the SQL, target data source, and query purpose before letting the Agent continue. sql-execute is limited to read-only DQL and metadata checks. SQL that can change data, schema, permissions, or transaction state must use SQL Task.

NineData returns the execution result or the next recommended action based on platform rules. Common outcomes include:

  • The query is executed and a result preview is returned.
  • Full SQL confirmation is required again.
  • The SQL is rejected by permission, rule, or safety policy.
  • The SQL must be handled through a SQL task or approval workflow.
  • Database execution fails, and an error message and request ID are returned.

Common decisions are EXECUTED, REJECTED, NEED_CONFIRMATION, NEED_SQL_TASK, and FAILED. A result preview may be limited by pagination or platform protection. Check the returned truncation state before treating the preview as complete.

Submit SQL Changes with Skill​

For SQL that changes data, schema, permissions, or transaction state, route the statement through a NineData SQL task. This includes INSERT, UPDATE, DELETE, CREATE, ALTER, DROP, GRANT, REVOKE, transaction control, potentially state-changing procedure calls, and mixed batches that contain any non-read-only statement.

The typical workflow is:

  1. Describe the change requirement in the AI Agent client.

    Use ninedata-skill to submit a SQL task for adding a remark column to the orders table in the test database.
  2. The Agent displays the target data source, database, full SQL, and optional rollback SQL.

  3. After confirming the content, explicitly agree to submit the SQL task.

  4. NineData creates the SQL task and returns the task ID, rule check result, approval status, and console task link.

  5. Continue to submit for approval, wait for approval, or execute the task based on the task status. For multi-step approval, process the current approval nodes until detail returns canExecute=true.

For approval, execution, pause, resume, stop, cancel, reject, or approval transfer operations, the Agent should present the current task status and operation impact, and continue only after your explicit confirmation.

Import Data with Skill​

Data Import writes data to a target database and remains subject to NineData permission checks, development policies, approval, execution controls, and audit logging.

  1. Prepare the import file and use Data Import to confirm the import type and file format supported by the target data source.
  2. In the AI Agent client, specify the target data source, database, schema, table, file, and reason for the import. The Agent should upload the local file and confirm that its status is available before submitting the task; a successful upload does not mean the data has been imported.
  3. The Agent should display the target location, file information, and import options. Review the import scope, overwrite risk, and rollback or cleanup plan, then explicitly agree to submit the Data Import task. If the platform flags sensitive data, you must explicitly confirm the warning; the Agent must not disable the sensitive-data check automatically.
  4. After NineData creates the task, the Agent should return the task ID, rule-check result, approval status, execution status, and console task link.
  5. If approval is required, process each current approval node returned by the platform. Execute the import only when the task permits execution and you explicitly authorize it.
  6. After the import finishes, check the task log, affected scope, and target-table data. If it fails, use the task error for troubleshooting instead of bypassing platform rules and resubmitting.

The upload result includes a file ID and file availability status. The import task result includes the task ID, target data source and table, import settings, and approval and execution status. In the returned information, log is the workflow operation history, while process-log is the import execution log.

Skill does not automatically approve or execute Data Import tasks. Before importing production data, confirm backup, approval, rollback, and test-data cleanup requirements.

Export Query Results with Skill​

Data Export generates a file from read-only query results. Supported file types include csv, excel, sql, json, xml, and txt. The workflow remains subject to rule checks, approval, execution permissions, and audit logging.

  1. Confirm the data source, database, schema, and complete read-only SQL.
  2. Submit the Data Export workflow and specify file type, charset, and export content when required.
  3. Query task detail and inspect rule checks, approval status, execution status, and fileAvailable.
  4. When approval is required, process the current approval node returned by the platform. Handle multi-step approval one node at a time.
  5. Execute the export only when canExecute=true and the user explicitly authorizes execution.
  6. Use the workflow log or process log to diagnose execution state.
  7. Request a download link only when detail returns fileAvailable=true and the user explicitly asks for it. Encryption-password retrieval requires separate authorization.

Submit, update, approval, execution, suspend, resume, stop, cancel, skip, download, and password retrieval are separate operations. Skill does not automatically approve a workflow, download a file, reveal an encryption password, or skip an export task.

Recommendations​

  • Before letting an AI Agent call Skill, confirm that the target data source, SQL content, and account permissions are correct.
  • For SQL that changes data or schema, use the SQL task capability to submit the SQL to an approval workflow.
  • For imports, confirm the file, target table, and affected scope first, then upload the file and confirm its availability before following Data Import approval and execution permissions.
  • For exports, confirm the SQL and file scope first, then follow Data Export approval, execution, and download permissions.
  • For read-only queries, execution plan retrieval, or development data validation, use the SQL execution capability within your organization permissions and security rules.
  • Skill results can help an AI Agent make decisions, but production changes should still be reviewed together with approval, audit, and business impact.
  • Do not let the Agent connect directly to databases, and do not store database usernames or passwords in the Skill.
  • Do not expose accessKeySecret, cookies, tokens, signatures, or database credentials in chats, logs, or commits.
  • When the platform returns rejection, failure, confirmation required, SQL task required, or approval required, do not bypass the platform decision.
  • Approval actions such as approve, reject, and transfer-approval are high-risk operations. Run them only after the user explicitly confirms the exact action.