Cloudflare Skill

Cloudflare Skill

Wrapper over the official Cloudflare CLI (cf, a Technical Preview — "the next version of Wrangler"). Typed operations for the core flows plus a custom passthrough that covers the entire cf surface. cf prints JSON to stdout by default, so results come back parsed in result.

Tool: cloudflare

Operations

Response

jsonCopy code
{
  "operation": "zones_list",
  "success": true,
  "result": [{ "id": "023e105f4ecef8ad9ca31a8372d0c353", "name": "example.com", "status": "active" }],
  "stdout": null,
  "stderr_tail": null
}

Parsed JSON lands in result; plain text (rare) lands in stdout. On failure the tool raises an error carrying cf's own message — surface it verbatim; cf's errors are precise (including "Not logged in", which tells the user exactly how to authenticate).

DNS records

record_body is the raw JSON request body (the API's DNS record shape — stable across cf versions, unlike per-field flags):

jsonCopy code
{ "operation": "dns_record_create", "zone": "example.com", "record_body": "{\"type\":\"A\",\"name\":\"www\",\"content\":\"192.0.2.1\",\"ttl\":1,\"proxied\":false}" }

Common record fields: type (A/AAAA/CNAME/TXT/MX/NS/SRV), name (record name; @ for the zone apex), content, ttl (1 = auto), proxied (orange-cloud), priority (MX/SRV). Get record ids from dns_records_list before deleting:

jsonCopy code
{ "operation": "dns_records_list", "zone": "example.com" }
{ "operation": "dns_record_delete", "zone": "example.com", "record_id": "023e105f4ecef8ad9ca31a8372d0c353" }

To update a record, use custom with --body:

jsonCopy code
{ "operation": "custom", "command": "dns records update <record-id> --zone example.com --body '{\"content\":\"192.0.2.2\"}'" }

Analytics — the GraphQL Analytics API

Cloudflare's legacy Zone Analytics REST API was deprecated (2021) and the DNS analytics REST API is scheduled for removal (2026-12-01); the official replacement for both is the GraphQL Analytics API. The cf CLI does not expose it (its commands are generated from the REST schema), so use the graphql_query operation — it POSTs to the official endpoint api.cloudflare.com/client/v4/graphql.

Requires the API token from Credentials -> Cloudflare (the OAuth login has no analytics scopes): permission Account > Account Analytics > Read, plus Zone > Analytics > Read for zone-scoped queries.

Zone HTTP traffic (official migration-guide query, current dataset):

jsonCopy code
{
  "operation": "graphql_query",
  "graphql_query": "query Sample($zoneTag: string, $start: Time, $end: Time) { viewer { zones(filter: {zoneTag: $zoneTag}) { series: httpRequestsAdaptiveGroups(limit: 5, orderBy: [count_DESC], filter: {datetime_geq: $start, datetime_lt: $end, requestSource: \"eyeball\"}) { count avg { sampleInterval } sum { visits edgeResponseBytes } dimensions { coloCode } } } } }",
  "graphql_variables": "{\"zoneTag\": \"<zone id>\", \"start\": \"2026-07-01T00:00:00Z\", \"end\": \"2026-07-15T00:00:00Z\"}"
}

Web Analytics / RUM (account-scoped datasets: rumPageloadEventsAdaptiveGroups, rumPerformanceEventsAdaptiveGroups, rumWebVitalsEventsAdaptiveGroups):

jsonCopy code
{
  "operation": "graphql_query",
  "graphql_query": "query Rum($accountTag: string, $start: Time, $end: Time) { viewer { accounts(filter: {accountTag: $accountTag}) { rumPageloadEventsAdaptiveGroups(limit: 100, orderBy: [count_DESC], filter: {datetime_geq: $start, datetime_lt: $end}) { count sum { visits } dimensions { requestPath } } } } }",
  "graphql_variables": "{\"accountTag\": \"<account id>\", \"start\": \"2026-07-01T00:00:00Z\", \"end\": \"2026-07-15T00:00:00Z\"}"
}

DNS analytics: dnsAnalyticsAdaptive / dnsAnalyticsAdaptiveGroups (zone- and account-scoped). Notes: a query spans at most 10 zones or 1 account; the user quota is 300 queries per 5 minutes; wide time ranges are sampled (multiply by sampleInterval for estimates). Web Analytics site CONFIG (create/list sites) lives on REST under accounts/{account_id}/rum/site_info — the cf CLI has no rum commands yet, and those endpoints also need the API token.

The full cf surface via custom

jsonCopy code
{ "operation": "custom", "command": "accounts list" }
{ "operation": "custom", "command": "dns records get <record-id> --zone example.com" }
{ "operation": "custom", "command": "registrar domains list" }
{ "operation": "custom", "command": "zones get --zone example.com" }
{ "operation": "custom", "command": "dns analytics report --zone example.com" }

Discovery helpers when unsure of a command's shape:

• { "operation": "custom", "command": "agent-context dns" } — cf's built-in per-product agent context (also agent-context --list for the product catalogue).
• { "operation": "custom", "command": "schema zones list" } — the API schema behind a command.
• Append --dry-run to any write command to validate without executing.

Quoting and escaping

command is parsed with shlex.split — wrap JSON bodies in single quotes: custom: "dns records create --zone example.com --body '{\"type\":\"A\",...}'".

Authentication

Two independent paths; the token wins when both exist (cf's documented resolution order):

1. OAuth login — Credentials Modal → Cloudflare → Login (cf opens the browser itself; localhost callback, so the browser must be on the server's machine), or cf auth login in a terminal. The OAuth grant is a FIXED scope set: zones, DNS (incl. dns_analytics:read), accounts, registrar, Workers — but NO Web Analytics/RUM or zone analytics scopes, and no way to request more.
2. API token or Global API Key — pasted in the credentials panel (or CLOUDFLARE_API_TOKEN / CLOUDFLARE_API_KEY+CLOUDFLARE_EMAIL in the server env; panel values take precedence). Required for graphql_query and anything else outside the OAuth scopes. Scoped token: dash.cloudflare.com/profile/api-tokens with the permission groups the task needs. Global API Key (cfk_ prefix): full access, needs the Account Email field filled alongside it.

If a command fails with an authentication or 403 error, tell the user which permission group the token needs; don't ask them to paste a token in chat.

Best practices

1. Run whoami first when auth state is unknown — it returns the authenticated email and scopes without side effects.
2. Pass zone as the domain name (cf resolves it to the zone ID).
3. Preview writes with --dry-run via custom when the user's intent is ambiguous.
4. Destructive operations (zone deletes, record deletes) — confirm with the user before running unless they explicitly asked.
5. Surface cf error messages verbatim — don't paraphrase.