A hand-curated, fidelity-verified specification and CLI for the Realtime Register REST API v2. Every non-SiteLock endpoint — 109 operations across 16 categories — is described in machine-readable YAML, reconciled field-by-field against the live HTML documentation at dm.realtimeregister.com/docs/api.

One package, two audiences:

• AI agents load SKILL.md + references/*.md as structured context and validate outgoing requests through rtr validate.
• Human developers use rtr at the terminal for schema-validated request construction, operation lookup, and doc-link auditing.

• Machine-verified fidelity — all 109 operations carry verified: docs, the full live-docs diff reports zero drifts, and a committed fingerprint lock (assets/spec/_fingerprints.json) turns the audit red if any verified contract is edited without re-verification. See docs/fidelity.md.
• The auth contract is part of the skill — every reference file and SKILL.md state it: Authorization: ApiKey <key>, never X-API-KEY, never Basic (deprecated upstream), never session keys.
• Runtime validation — JSON Schemas derived from the YAML, evaluated by ajv with format checks for email, uri, date, date-time, ipv4, ipv6. rtr describe <op> --format json emits the same schemas for programmatic consumers.
• Agent-first docs — SKILL.md + references/<category>.md are the only files an LLM needs to load; they are terse, structured, and stable.
• Zero-surprise CLI — six subcommands, no config files, no env vars, no network access except doctor and scrape.
• CamelCase everywhere — matches the official TypeScript SDK and the on-the-wire protocol.
• 136 tests guard the pipeline: an Ajv compile gate across every operation, a byte-exact sync check between the YAML and the generated references, CLI behavior tests, and scraper fixtures.

Drop the skill (SKILL.md + references/ + assets/spec/) into your agent client's skills directory — one command, no prior install:

npx @cave-man/realtime-register-skills install

The installer auto-detects these targets. When multiple matches exist and stdin is a TTY, you'll be prompted to choose one; non-interactive callers (CI, --yes, piped input) get the first match:

Tools with no skills mechanism get a pointer file via --pointer instead — a small sentinel-delimited block telling the agent where the skill lives:

npx @cave-man/realtime-register-skills install                      # auto-detected skill target
npx @cave-man/realtime-register-skills install --pointer all        # + every pointer file
npx @cave-man/realtime-register-skills install --pointer junie,agentsmd

Pointer writes only ever touch the sentinel-delimited block they own; if the target file already has unrelated content and no sentinel yet, the installer prints the block instead of writing it, so you can add it by hand.

Verified against official docs on 2026-07-04: developers.openai.com/codex/skills, geminicli.com/docs/cli/skills/, docs.augmentcode.com/cli/skills, junie.jetbrains.com/docs/guidelines-and-memory.html, docs.github.com, cursor.com/docs/context/rules, agents.md. ~/.codex/skills/ is a best-effort target — see github.com/openai/codex/issues/14337.

Override with --target <dir> or the REALTIME_REGISTER_SKILL_DIR env var. Installer subcommands:

npx @cave-man/realtime-register-skills install --dry-run           # preview
npx @cave-man/realtime-register-skills install --target ./skills   # explicit dir
npx @cave-man/realtime-register-skills install --global            # + npm i -g (rtr on PATH)
npx @cave-man/realtime-register-skills where                       # list detected targets
npx @cave-man/realtime-register-skills uninstall                   # remove the skill
npx @cave-man/realtime-register-skills uninstall --pointer all     # + remove pointer blocks

# one-off (recommended)
npx @cave-man/realtime-register-skills rtr --help

# or globally
npm install -g @cave-man/realtime-register-skills
rtr --help

rtr describe createDomain                 # human-readable contract
rtr describe createDomain --format json   # machine-readable: op + auth + JSON Schemas

cat > /tmp/register.json <<'JSON'
{
  "registrant": "H1234567",
  "period": 12,
  "ns": ["ns1.example.com", "ns2.example.com"],
  "privacyProtect": true
}
JSON

rtr validate createDomain --body /tmp/register.json
# → exits 0 if the body matches the schema, 1 otherwise

rtr doctor
# OK  200 createDomain      https://dm.realtimeregister.com/docs/api/domains/create
# OK  200 getDomain         https://dm.realtimeregister.com/docs/api/domains/get
# … 109 lines total, non-200s exit non-zero.

Two binaries ship in this package:

skills — the installer entry point (npx @cave-man/realtime-register-skills …).

rtr — the operation / schema CLI.

Full flag reference in docs/cli.md.

The source of truth is assets/spec/:

assets/spec/
├── _shared.yaml        # enums and reusable nested types
├── domains.yaml        # 13 operations (11 customer + 2 gateway)
├── hosts.yaml          #  5 operations
├── contacts.yaml       # 11 operations
├── dnsZones.yaml       #  9 operations
├── dnsTemplates.yaml   #  5 operations
├── validation.yaml     #  2 operations
├── ssl.yaml            # 17 operations
├── sslAcme.yaml        #  7 operations
├── notifications.yaml  #  5 operations
├── processes.yaml      #  5 operations
├── customers.yaml      #  2 operations
├── brands.yaml         # 10 operations (CRUD + templates + locales)
├── financial.yaml      #  4 operations
├── tlds.yaml           #  2 operations
├── providers.yaml      #  7 operations (incl. gateway-only registry accounts)
└── misc.yaml           #  5 operations (IsProxy + 4 ADAC WebSocket actions)

Each operation carries operationId, method, path, docUrl, async, authScope, deprecated, verified, liveDiff, pathParams, queryParams, requestBody.fields, responses, errors, gotchas, and examples. Full schema in docs/spec-format.md.

The SKILL.md file at the repo root is the canonical entry point. Load it into your agent's system context along with the relevant references/<category>.md file(s). A typical flow:

1. Agent receives user request ("renew example.com for two years").
2. Agent loads SKILL.md → learns the hard rules (auth is Authorization: ApiKey <key> — never X-API-KEY, never Basic; camelCase; period in months; async polling; billable acknowledgment).
3. Agent loads references/domains.md → finds the renewDomain entry (every reference file opens with the Authentication section).
4. Agent builds the payload and calls rtr validate renewDomain --body … before issuing the HTTP request — or reads the machine contract via rtr describe renewDomain --format json.
5. On HTTP 202, agent polls /v2/processes/{processId} via the processes category.

End-to-end integration notes, a programmatic validation snippet, and the billable-acknowledgment / async-mutation patterns are documented in docs/agent-integration.md.

Every operation declares a verified marker:

• docs — path, URL parameters, query parameters, and request-content fields reconciled against the live HTML.
• sdk — derived from the public TypeScript SDK and the navigation slug only; no HTML-level reconciliation yet.

As of v0.3.0, all 109 operations carry verified: docs, re-verified field-by-field against a full crawl of the live documentation — and the claim is machine-enforced from here on: a committed fingerprint lock (assets/spec/_fingerprints.json) fails the audit if a verified contract is edited without re-verification, and the weekly drift job diffs every operation's method, path, and required fields against the live pages (non-machine-readable doc pages are marked liveDiff: false). Promotion workflow, drift policy, and enforcement tooling are documented in docs/fidelity.md.

• Node.js 20.11 or newer.
• No runtime Realtime Register account needed for the CLI — schema validation is local. You only need credentials to actually call api.yoursrs.com.

See CONTRIBUTING.md. In short: never edit references/*.md directly (they are generated), keep field names camelCase, and run npm run verify — the one-command build + lint + test + audit gate — before opening a PR.

MIT. Not affiliated with Realtime Register B.V.; this project merely consumes their public documentation.