Prompt: file a support ticket with ird that gets answered first time

Have an agent turn "it doesn't work" into a ticket the ird team can act on: reproduce the problem, collect the right diagnostics with the CLI, redact anything secret, pick the service, keep it inside the size limits, submit it only after you read it, and follow the reply thread. Support is free per the price book.

Needs: ird access (the ird CLI logged in, or the MCP server); the failing command or tool call Costs: none per the price book — support.writes and support.reads are zero in credits_prices; the diagnostic reads bill their own *.reads keys (zero or negligible — read them there) Group: housekeeping

Paste everything below the line into a fresh conversation. Fill in the FILL IN block first.

Task

Investigate the problem below, assemble a ticket, show it to me, and submit it with support_create_ticket after I approve the exact text. Then check for replies only when I ask.

FILL IN

PROBLEM:   ird dns set returns "zone not found" for example.ro since this morning   # one sentence
RESOURCE:  example.ro                     # the domain / hostname / id involved, or blank
WHEN:      2026-09-09 ~07:30 UTC          # first noticed
TRIED:     re-login, ird update           # what I already did

Non-negotiable rules

1. Nothing leaves my machine before I approve. Show the full title and message; submit only after an explicit yes. The message is sent on my behalf.
2. Redact. Never paste an API key (--api-key, IRD_API_KEY, ird keys output), TURN or mailbox credentials, one-time root passwords, private keys, or third parties' personal data. Replace with <redacted>; keep resource ids, hostnames and timestamps — those are what the team needs.
3. Limits. Title ≤ 120 characters, message ≤ 5000 (from support_create_ticket). Overflow goes into a first reply, not a second ticket.
4. One ticket per problem. Check support_list_tickets first; if an open ticket covers it, support_reply there instead.
5. Read-only diagnostics. Only *_list, *_get, status, whoami, version and re-running the failing command. Nothing that creates, changes or bills.

Steps

A. Reproduce. Run the failing command once more with --json (CLI) or repeat the tool call, and keep the exact input and the verbatim error, including any error code or request id in the output. Note the UTC time of the run (date -u). If it now works, say so — the ticket becomes "happened at WHEN, resolved by itself".

B. Environment. ird version; ird whoami --json (account, not key); ird status --json for the account overview. With the MCP server note the MCP client and that the CLI was not used.

C. Resource state. The read that shows the resource as the API sees it, with ids: domains → list_domains and dns_get_zone(domain) (ird domains list --json, ird dns records <domain> --json); claims → dns_claim_get; IPFS → ipfs_pins, ipns_list, ipfs_domains; tunnels → list_tunnels (ird tunnels status <host> and ird tunnels logs are CLI-only); servers → vps_list; apps → apps_get; p2p → p2p_list_apps, p2p_usage, GET https://<host>/.well-known/ap2p; billing → credits_balance, ird credits ledger (CLI-only). Include only the entries that matter, trimmed, with id, state and timestamps.

D. Service. domains for anything about a .ro registration, transfer, renewal or registry contact; general for DNS records, IPFS, tunnels, servers, apps, p2p, billing and the CLI/MCP itself; consultancy or spatial only if that is what I bought. Default is general.

E. Compose. Title: <service area>: <symptom> (<resource>), ≤ 120 chars. Message, in this order: what I did (exact command/tool call); what I expected; what happened (verbatim error); resource ids; times in UTC; environment (B); what I TRIED; the trimmed diagnostics (C). Count characters; if over 5000, keep the error and ids, move raw listings to a follow-up reply. Run the redaction pass (rule 2) and show me both fields.

F. Submit. After my yes: support_create_ticket(title=…, message=…, service=…) (CLI: write the message to a file, then ird support new -s "<title>" --service <service> --file report.txt). Keep the returned ticket id. The team answers asynchronously — replies also arrive by email, and answering that email lands on the same ticket.

G. Follow up — no polling. When I ask: support_get_ticket(ticket_id) (CLI: ird support show <id>) shows the whole thread and marks replies seen; support_list_tickets (ird support list) shows unread counts. Answer questions from the team with support_reply(ticket_id, message=…) (ird support reply <id> -m … or --file), same limits, same redaction. When it is fixed, reply once to say so.

Deliverable

The ticket id, service, title, and the submitted message as sent; a list of what was redacted (by kind, not value); the diagnostics kept locally in full for later replies; and the one-line command to check for a reply.