service-de-channel-create

Inserting Messaging Channels

Reference File Index

What this skill does

Given a {MESSAGE_TYPE} and its inputs, creates a MessagingChannel record with PlatformType=Enhanced, IsActive=false. The same server-side path (LiveMessageSetupServiceImpl.addChannel → LiveMessageChannelsUtil.findOrCreateCsotMessagingChannel) also inserts a MessagingChannelUsage row (DeploymentType=Conversation, DeploymentStatus=New on fresh insert — Disabled on a channel that was previously activated then deactivated) — we get it for free.

One primary path for every type: Connect REST API. POST /services/data/v{V}/connect/livemessage/channels with an org OAuth session managed by sf api request rest. Public, sanctioned, live since API v66+. No Aura session, no token extracted into shell state.

Idempotent on (MessagingPlatformKey, MessageType) — re-firing with the same platform key returns the existing channel's id. This matters because MessagingChannel records are not deletable via standard means, so the preflight (Stage 1) is the practical idempotency guard.

All types return the same envelope shape, so callers don't branch on type after insert: {ok, channelId, mcuId, developerName, isActive:false, messageType, messagingPlatformKey, path, created, durationMs}.

Supported types (the only branch point)

Unsupported (would need their own leaf logic, not yet built): WeChat, MsCopilot, Alexa. For these, emit the unsupported-type envelope (Stage 0).

When NOT to use this skill

• End-to-end setup. If the user wants insert → route → activate, invoke

service-de-headless-channel-configure — it orchestrates this skill plus routing, consent, and activation. This skill only handles the insertion step (see Stage 0).

• The channel already exists and you only want to re-route / activate it. Skip insertion;

call service-de-channel-routing-configure / service-de-channel-activate with the existing channelId.

Inputs (from caller)

• {MESSAGE_TYPE} — one of WhatsApp, Line, AppleBusinessChat, Facebook, Text.

Case-sensitive (matches the SF picklist). Text is the enum value for SMS — there is no SMS type.

• {ORG_ALIAS} — optional; sf CLI target-org alias. Default: sf config get target-org.
• {API_VERSION} — optional; Connect API version. Default: the org's current API version

(sf org display --json | jq -r '.result.apiVersion'). Always use the org's native version to avoid 404s.

• {CHANNEL_NAME} — display name shown in Setup → Messaging Settings. Required for

AppleBusinessChat (no phone/screen-name fallback); prompted if omitted for others.

Type-specific inputs:

• WhatsApp: {WABA_ID} (sent as externalAccountId), {PHONE_NUMBER_ID} (the platform

key — must differ from WABA_ID), {ISO_COUNTRY_CODE} (optional, default US).

• Line: {LINE_CHANNEL_ID} (platform key), {LINE_CHANNEL_SECRET}, {LINE_ACCESS_TOKEN},

{VERIFY_TOKEN_FIRST} (optional, default true).

• AppleBusinessChat: {APPLE_BC_ID} (platform key, GUID).
• Facebook: {PAGE_ID} (optional — OAuth fetches/prompts if absent).
• Text (SMS): {SMS_NUMBER} (the platform key — the phone number / short code string, already

provisioned to the org), {SMS_PROVIDER} (sent as smsProvider — the smsProvider key must be present in the body or the insert 400s with "SMS provider is missing", but its value is not validated or stored, so an empty string "" works; default to "" when the caller doesn't supply a provider), {ISO_COUNTRY_CODE} (optional, default US). This skill does not provision the number — the caller must already know their number.

Output (to caller)

Success:

{"ok": true, "channelId": "0Mj...", "mcuId": "0gL...", "developerName": "...",
 "isActive": false, "messageType": "{MESSAGE_TYPE}", "messagingPlatformKey": "...",
 "path": "connect" | "preflight", "created": true | false, "durationMs": 1234}

created: false + path: "preflight" means the row already existed and was returned idempotently.

Failure (shared):

{"ok": false, "kind": "wrong-skill", "correctSkill": "service-de-headless-channel-configure", "hint": "..."}
{"ok": false, "kind": "unsupported-type", "supportedTypes": ["WhatsApp","Line","AppleBusinessChat","Facebook","Text"], "hint": "..."}
{"ok": false, "kind": "missing-input", "missing": ["..."], "hint": "caller must provide all required inputs for this type"}
{"ok": false, "kind": "auth",     "hint": "OAuth token invalid / expired — run 'sf org login web'"}
{"ok": false, "kind": "business", "message": "...", "errorCode": "..."}
{"ok": false, "kind": "transport","status": 500, "message": "..."}
{"ok": false, "kind": "verify-failed", "hint": "Connect returned 201 but follow-up SOQL shows no matching channel"}

Type-specific failure kinds (partnership-blocked, user-declined, waba-manage-failed, meta-precondition for WhatsApp; line-token-invalid for LINE; apple-precondition for Apple; oauth-not-complete, oauth-failed for Facebook) are documented in the per-type reference files.

Stage 0: Enforce orchestrator-only invocation

This skill is INTERNAL USE ONLY. It should only be invoked by the service-de-headless-channel-configure orchestrator.

If the user's original request was to "install", "set up", "create", or "activate" a messaging channel (or any phrasing that implies end-to-end setup), immediately return:

{
  "ok": false,
  "kind": "wrong-skill",
  "hint": "This is an internal insertion skill. For end-to-end channel setup (insert → route → activate), invoke 'service-de-headless-channel-configure' instead with the desired MESSAGE_TYPE.",
  "correctSkill": "service-de-headless-channel-configure"
}

And render to the user:

Error: Wrong skill invoked.

This skill (service-de-channel-create) only handles the insertion step.
For complete channel setup, use: /service-de-headless-channel-configure

That orchestrator runs: insert → configure routing → activate in one flow.

Only proceed to Stage 0.1 if:

• This skill was invoked programmatically by service-de-headless-channel-configure
• The user explicitly stated they ONLY want insertion (not routing or activation)
• This is a retry/resume scenario where routing/activation already succeeded

When in doubt, redirect to the orchestrator.

Stage 0.1: Validate type and required inputs

Look up {MESSAGE_TYPE} in the supported-types table. If absent, emit the unsupported-type envelope and return.

Validate the type's required inputs are present (Facebook is the exception — no required inputs; OAuth → page selection → prompts happen inside the Facebook flow). If any are missing, emit missing-input with the list and return.

WhatsApp only: additionally assert {WABA_ID} != {PHONE_NUMBER_ID} — they are distinct Meta entities and reusing one value for both causes cryptic activation failures. If equal, emit {ok:false, kind:"invalid-input", hint:"WABA_ID and PHONE_NUMBER_ID must differ — WABA_ID is the WhatsApp Business Account ID; PHONE_NUMBER_ID is a specific phone number within it. Find both in Meta Business Manager → WhatsApp Accounts."}.

Capture {T0} = Date.now() for durationMs.

Stage 0.2: Establish the scratch directory

Every transient artifact this skill writes (preflight query results, request bodies, Connect responses) goes into a per-run scratch directory rooted inside the harness working area — never a bare absolute /tmp path, which is world-writable and non-portable. Establish it once:

SCRATCH_DIR="$(mktemp -d "${outputDir:-${TMPDIR:-/tmp}}/icc.XXXXXX")"

The base dir is the harness-provided ${outputDir} when set, else the private user-scoped ${TMPDIR}, else /tmp only as a last resort. All later stages (and the reference flows) write under ${SCRATCH_DIR}/; the durable report stays under ${outputDir}.

Stage 1: Preflight — is the channel already here?

SOQL-probe for an existing row first; a hit lets us skip all prerequisites and the insert.

For Facebook without a `{PAGE_ID}`, skip this stage — the page id isn't known until OAuth → page selection completes (see references/facebook.md). Run the preflight after page selection.

Two queries, not a subquery. The (SELECT ... FROM MessagingChannelUsages) child subquery fails with INVALID_TYPE on orgs where the child relationship is unnameable (verified on wadtesting / PRODDEBUG). The FK-keyed form works everywhere.

sf data query --target-org '{ORG_ALIAS}' \
  --query "SELECT Id, DeveloperName, MasterLabel, IsActive, MessagingPlatformKey FROM MessagingChannel WHERE MessagingPlatformKey = '{PLATFORM_KEY}' AND MessageType = '{MESSAGE_TYPE}'" \
  --json > "${SCRATCH_DIR}/preflight.json"

If records.length === 1:

• Record {CHANNEL_ID}, {DEVELOPER_NAME}, {IS_ACTIVE}.
• Second query for the MCU by FK:

  sf data query --target-org '{ORG_ALIAS}' \
    --query "SELECT Id, DeploymentStatus FROM MessagingChannelUsage WHERE MessagingChannelId = '{CHANNEL_ID}'" \
    --json > "${SCRATCH_DIR}/preflight-mcu.json"

Record {MCU_ID} (empty MCU: flag, don't fail — the activation skill handles it).

• Emit the success envelope with path: "preflight", created: false. Return.

If records.length > 1: ambiguous (uniqueness should prevent this) — proceed with the first, note a warning in the envelope.

If records.length === 0: proceed to Stage 2.

Stage 2-3: Type-specific prerequisites

Most types have no prerequisite here — go straight to Stage 4.

• WhatsApp: requires the Meta WABA↔Salesforce partnership (Stage 2, via

service-de-waba-integrate) and the SF-side webhook subscription + credit-line share (Stage 3, POST /connect/livemessage/whatsapp/business-account). Load `references/whatsapp.md` and follow it before inserting.

• Line: optional token preflight against https://api.line.me/v2/bot/info if

{VERIFY_TOKEN_FIRST} is true. See `references/line.md`.

• Facebook: browser-based OAuth authorization + page-list fetch + page selection happen here,

before the insert. Load `references/facebook.md` and follow it.

• Apple: none. (Apple's certificate handshake is provisioned on Apple's side and is per-message

at runtime, not at channel insert — see references/apple.md.)

• Text (SMS): none at insert. The number must already be provisioned to the org and the caller

must know its smsProvider, but this skill performs no provisioning step — it inserts the record for a number the caller already has. (ESP registration runs at activation, not insert — see references/sms.md.)

Stage 4: Insert via Connect REST API

Shared across all types. The only per-type difference is the request body; the authProviderId lookup, the POST invocation, and the response-classification table are identical.

Stage 4.1: authProviderId (org-global; required for every type except Text)

sf data query --target-org '{ORG_ALIAS}' \
  --query "SELECT Id FROM AuthProvider WHERE DeveloperName = 'LiveMessageSetup'" \
  --json > "${SCRATCH_DIR}/authprovider.json"

One row → {AUTH_PROVIDER_ID}. Zero rows → branch on type:

• `Text` — non-fatal. Text's Connect signature doesn't take authProviderId, so leave

{AUTH_PROVIDER_ID} unset and continue to Stage 4.2; the POST omits the parameter entirely (see below). Do not emit the business error for Text — an org that has never run LiveMessage setup is exactly where the Text-only path matters, and blocking here would make the advertised SMS insert unreachable on those orgs.

• every other type (WhatsApp, Line, AppleBusinessChat, Facebook) — the org has never run

LiveMessage setup and the insert can't proceed; emit {ok:false, kind:"business", message:"No AuthProvider 'LiveMessageSetup' found — run the Messaging Setup wizard once on this org first"}.

The Connect endpoint rejects the POST with 400 ILLEGAL_QUERY_PARAMETER_VALUE "Missing argument authProviderId" if this URL parameter is absent — verified for every type except `Text`, Apple included. (Text has its own Connect signature that doesn't require it.) When the lookup finds a row, send authProviderId for every type including Text (harmless there); when it finds none, only Text may proceed, and it proceeds without the parameter. authProviderId goes in the URL query string, never the body (body → 400 JSON_PARSER_ERROR "Unrecognized field 'authProviderId'").

Stage 4.2: Build the type-specific body and POST

Resolve {CHANNEL_NAME} first (prompt if required and absent — see per-type ref). Then build the body from the supported-types table and POST. Write the body to a file (--body @...) — LINE tokens are long and can contain shell-hostile characters.

Append the authProviderId query parameter only when Stage 4.1 resolved one — otherwise (the Text-with-no-AuthProvider case) POST to the bare endpoint, since an empty ?authProviderId= would trip the same ILLEGAL_QUERY_PARAMETER_VALUE:

URL="/services/data/v{API_VERSION}/connect/livemessage/channels"
[ -n "${AUTH_PROVIDER_ID:-}" ] && URL="${URL}?authProviderId=${AUTH_PROVIDER_ID}"

sf api request rest "${URL}" \
  --method POST --target-org '{ORG_ALIAS}' \
  --header 'Content-Type: application/json' --header 'Accept: application/json' \
  --body @"${SCRATCH_DIR}/body.json" --include > "${SCRATCH_DIR}/connect-response.txt" 2>&1

--include prints the HTTP status/header block before the body — read the status from there, not a trailing -w marker. sf api request rest manages OAuth internally.

For the exact per-type body shape and the full response-classification table (including the `DUPLICATE_VALUE` race back to Stage 1, the Facebook `AUTH_PROVIDER_NEEDS_AUTH` 401-vs-400 disambiguation, and per-type precondition errors), load `references/connect-insert.md` plus the per-type ref.

Stage 5: Verify the insert

Two queries (same no-subquery reason as Stage 1). Keep to the universal field set — ExternalAccountId and MessagingPlatform don't exist on all orgs.

sf data query --target-org '{ORG_ALIAS}' \
  --query "SELECT Id, DeveloperName, MasterLabel, MessageType, IsActive, MessagingPlatformKey FROM MessagingChannel WHERE Id = '{CHANNEL_ID}'" --json

sf data query --target-org '{ORG_ALIAS}' \
  --query "SELECT Id, DeploymentStatus, DeploymentType FROM MessagingChannelUsage WHERE MessagingChannelId = '{CHANNEL_ID}'" --json

Expected: channel IsActive=false; exactly one MCU, DeploymentType=Conversation, DeploymentStatus in {New, Disabled} (fresh inserts land in New; Disabled is a previously-activated-then-deactivated channel — both are valid starting states for activation). Empty MCU query → {ok:false, kind:"verify-failed", hint:"Connect returned {CHANNEL_ID} but no child MessagingChannelUsage was created"}.

Trust the DB, not the Connect response body, for MasterLabel / DeveloperName — the 201 body carries stale input-rep projections (see per-type gotchas).

Stage 6: Report to caller

{"ok": true, "channelId": "{CHANNEL_ID}", "mcuId": "<from Stage 5>",
 "developerName": "<authoritative, from Stage 5>", "isActive": false,
 "messageType": "{MESSAGE_TYPE}", "messagingPlatformKey": "{PLATFORM_KEY}",
 "path": "connect", "created": true, "durationMs": <Date.now() - T0>}

Rendered (when invoked directly rather than by the orchestrator):

• Success — MessagingChannel {CHANNEL_ID} ({developerName}) created via {path}. MCU {mcuId} auto-created. IsActive=false — run routing, then consent, then activate next.
• Info: Channel already exists — {CHANNEL_ID} ({developerName}). Proceed to routing/activation.
• Error: {kind}: {message or hint}

On success the caller can immediately invoke service-de-channel-routing-configure with the returned channelId, then service-de-channel-consent-configure, then service-de-channel-activate. Those three are message-type-agnostic (they operate on MessagingChannel.Id + MessagingChannelUsage.Id), so there's no downstream dispatch concern.

Gotchas (shared)

1. Case-sensitive type names. "whatsapp" ≠ "WhatsApp" — the SF picklist, SOQL filters,

and Connect enum all match exactly. Normalize upstream; don't make Stage 0.1 case-insensitive.

1. Preflight is the idempotency guard. MessagingChannel rows aren't deletable via standard

means, so re-runs rely on Stage 1 finding the existing row.

1. Envelope shape is a tight coupling with callers. If a field is renamed here, every caller

(especially service-de-headless-channel-configure) silently sees the change — break loudly in tests rather than remap field names.

1. Per-type gotchas live in the reference files. WABA/phone-id confusion, LINE long-lived vs

short-lived tokens, Apple's empty-config body + no-provisioning-callout, Facebook OAuth page selection + 401 disambiguation — load the matching references/<type>.md when debugging that type.