msgbyte/tianji/tianji-data-query

Tianji Analytics

Query any read-only data from the Tianji monitoring and analytics platform.

Configuration

Three values are required (provided via skill config):

Making API Requests

Before querying data, fetch and read the target instance's OpenAPI document. TIANJI_BASE_URL is the instance URL without the /open suffix:

curl --fail-with-body --silent --show-error \
  "${TIANJI_BASE_URL%/}/open/_document"

Confirm a successful response containing an OpenAPI JSON object with openapi and paths, not HTML or an error payload. If discovery fails, report the failure and stop API queries; do not substitute bundled paths or guess endpoints. The instance must enable OpenAPI to serve this document.

Use the live document to confirm the operation's method, path, path/query parameters, response schema, and authentication. Resolve referenced schemas ($ref), including required fields, types, formats, defaults, enums, and pagination. Resolve servers against the target instance before appending an operation path (normally /open; do not add it twice). Keep authenticated requests on the intended instance and send the API key via Authorization: Bearer ..., never in the URL or output.

Only GET requests are allowed, even though the live document also describes writes. If the requested operation has no exported GET route, report that limitation. Do not derive /open routes from dashboard tRPC procedures; those need separate maintained references and are outside this skill.

The target document takes precedence over bundled references and examples. Reuse it within the same task and target; fetch again after switching instances, an upgrade, or a schema-related failure. Parse successful API responses as JSON and report HTTP/API errors rather than treating them as query results.

Service Domains

api-endpoints.md and openapi-readonly.json are bundled snapshots for discovering likely operations. Their endpoint counts and schemas may differ from the target; confirm every selected operation against its live document.

Workflow

1. Identify the service domain from the user's question
2. Read the target's /open/_document and find the relevant exported GET operation; bundled references are lookup hints only
3. Construct the GET request using the live path and parameter schemas, actual resource IDs, and required authentication
4. Parse the JSON response, redact sensitive fields, and summarize for the user

Common Scenarios

The routes below illustrate the bundled version. Confirm paths, parameters, timestamp formats, and pagination against the target document before using them.

Website traffic overview

GET /open/workspace/{workspaceId}/website/all

Pick the target website ID, then:

GET /open/workspace/{workspaceId}/website/{websiteId}/stats?startAt={timestamp}&endAt={timestamp}

Monitor health check

GET /open/workspace/{workspaceId}/monitor/all

Pick the target monitor ID, then:

GET /open/workspace/{workspaceId}/monitor/{monitorId}/get
GET /open/workspace/{workspaceId}/monitor/{monitorId}/status

Survey results analysis

GET /open/workspace/{workspaceId}/survey/all

Pick the target survey ID, then:

GET /open/workspace/{workspaceId}/survey/{surveyId}/result/list?startAt={timestamp}&endAt={timestamp}&limit=50
GET /open/workspace/{workspaceId}/survey/{surveyId}/stats?startAt={timestamp}&endAt={timestamp}

Feed event inspection

GET /open/workspace/{workspaceId}/feed/channels

Pick the channel ID, then:

GET /open/workspace/{workspaceId}/feed/{channelId}/fetchEventsByCursor?limit=20

Sensitive Data Handling

Some GET endpoints may return fields containing platform-stored secrets (e.g. modelApiKey, customModelBaseUrl in AI Gateway responses). Additionally, endpoints like workspace members, audit logs, and billing may contain PII or internal details.

Rules:

• Live schemas and API responses are not redacted by the bundled schema's filtering; always apply these rules yourself
• NEVER display modelApiKey, apiKey, secret, token, password, or credential fields to the user
• Redact or omit these fields when summarizing API responses
• When querying workspace members or audit logs, only surface non-sensitive metadata (names, roles, timestamps) unless the user explicitly requests full detail

Notes

• Use timestamp formats and metric type enums from the target operation; bundled examples use milliseconds since epoch
• Follow the target operation's pagination contract; some endpoints use cursor and return nextCursor