n8n-nodes-sendrepute

An n8n community package for SendRepute customer services. Its original classification node
accepts only sender, subject, and body; it has no recipient or attachment
input.

Customer services

The additional SendRepute Services node exposes the current non-deprecated
customer API operations: account, billing, catalogs, AI rewrites, standard and VIP
templates, builder services, and campaign insights. Choose an operation to reveal
its contract fields. Structured documents use JSON fields.

Every non-GET operation requires explicit authorization for each input item.
Fetch and review the appropriate price quote before enabling that authorization
and entering its expected price. Requests are not automatically retried. Keep
recovery identifiers and use result/recovery operations after ambiguous failures.
API key scopes, membership eligibility and spending limits still apply.

The original classification node and its workflow decisions remain unchanged.
The package is available on npm; this does not imply approval for n8n Cloud's
verified community catalog.

Security and decision model

• Paid-analysis consent is explicit and defaults to off. With consent off,
  no request is made.
• The credential uses n8n's password-field credential schema and is sent only
  as a bearer header. The node never includes it in output or errors.
• Requests go only to the fixed https://www.sendrepute.com/api/v1/classify
  endpoint, use verified platform TLS, have a 20-second timeout, reject
  redirects, and cap response bodies at 1 MiB.
• The node performs exactly one request per consenting input item. It does not
  retry paid analysis. The service may return a cached receipt for an exact
  repeat, but callers must not assume a request is free.
• A score must be a finite number from 0 through 1. Invalid JSON, malformed
  results, network failures, authentication/authorization failures, insufficient
  balance, and rate limiting are failures—not spam classifications.
• Advisory mode always allows a successfully analyzed item and emits a
  separate recommendation. Blocking mode blocks when the probability is
  at or above the threshold.
• Fail closed returns decision: "block" when analysis cannot run. Fail
  open returns decision: "allow" plus analysisStatus: "failed" and a
  failure category. Consumers must check allowed; merely receiving output is
  not an allow decision.

Classification is a pre-send content safety signal. It does not guarantee
inbox placement, email delivery, or compliance.

Build and test reproducibly

Requirements: Node.js 20.19 or newer and npm. From this directory:

git clone https://github.com/sendrepute/n8n-nodes-sendrepute.git
cd n8n-nodes-sendrepute
npm ci
npm run pack:check
npm pack --ignore-scripts

npm ci uses this package's committed package-lock.json; it does not modify
the repository root lockfile. pack:check compiles against the real published
n8n-workflow 2.39.3 types and runtime, runs offline HTTP transport and node tests, checks
tarball contents, and installs that tarball into an isolated temporary npm
project. TypeScript's skipLibCheck is enabled because the
published n8n type bundle references optional packages that are not required by
this node; this package's own source remains strict-checked.

Tests make no paid API calls and send no email. Their HTTP server is bound to
loopback only.

The tested dependency is pinned to the registry's stable release, 2.39.3,
rather than its older latest alias (2.16.0) or beta (2.41.0). The peer range
is >=2.39.3 <3; use a patched self-hosted n8n release that supplies a compatible
workflow runtime. Do not force-install into an older incompatible host.
Run npm audit after npm ci to check current advisory data. The upgraded
lockfile passed a full audit with zero reported vulnerabilities when tested;
this is a point-in-time result, not a guarantee against future advisories.

Install on self-hosted n8n

1. Build the tarball using the commands above.
2. On the n8n host, create the custom-node directory if necessary:
   mkdir -p ~/.n8n/nodes.
3. In that directory, run
   npm install /absolute/path/to/n8n-nodes-sendrepute-0.1.0.tgz.
4. Restart the self-hosted n8n process and confirm SendRepute appears in the
   node picker.
5. Create a SendRepute API credential containing a server-side customer API
   key with the classify scope. Do not put the key in workflow JSON,
   expressions, environment output, or logs.

Because the package is unpublished, n8n's Settings → Community nodes npm
name installation and n8n Cloud community-node installation are not supported.

Workflow use

Import examples/classify-email.workflow.json. It intentionally contains no
credential reference or secret and keeps paid consent disabled. Attach a
credential, review the inputs, threshold, decision mode, and failure policy,
then deliberately enable paid consent.

On completion the node preserves the input JSON and replaces the complete
sendrepute namespace with an explicit analysisStatus, decision, allowed,
recommendation, validated result fields, a billing receipt, and the advisory
disclaimer. On failure it replaces that namespace with a result that omits
classification labels/scores and contains only a sanitized failureCategory
and errorCode. This prevents stale decision fields from a previous run from
being mistaken for the current result. The example routes the result through an
IF node and leaves the false output unconnected, so only
sendrepute.allowed === true continues.

Supported hosted route: POST /api/v1/classify (the customer contract's
/v1/classify behind the hosted /api prefix), with sender (1–320
characters), subject (1–998), and body (1–524,288). Model selection is
intentionally left to the account preference. Recipients, attachments,
rewrites, sending, arbitrary origins, custom headers, and automatic retries are
not supported.

Complete node parameter reference

The node type is n8n-nodes-sendrepute.sendRepute, version 1. It has one main
input and one main output. It processes input items sequentially, preserving
item pairing. Configure these properties in the node editor:

Credential type sendReputeApi has one required password field: apiKey.
Credential lookup failure or a non-string key stops node execution before the
item loop. Empty, oversized or newline-containing string credentials are
reported as per-item authentication failures when paid analysis is attempted.

For an upstream item containing sender, subject and body, use expressions
={{ $json.sender }}, ={{ $json.subject }}, and ={{ $json.body }}. The included
manual-trigger workflow does not generate message fields: add an Edit Fields
node or your actual source before classification. Keep consent off until the
input and billing policy have been reviewed.

Output and failure reference

The node copies input JSON, replacing any existing sendrepute object:

• Successful analysis: analysisStatus: "completed", decision (allow or
  block), Boolean allowed, recommendation, decisionMode, threshold,
  label (inbox or spam), spamProbability, confidence (low, medium,
  high), reasons (objects with signal, detail, finite weight),
  flaggedTerms, analyzedFields, model, modelVersion, analyzedAt,
  requestId, and billing (chargedMillicents, replayed).
• Known failure: analysisStatus: "failed", decision, allowed,
  failurePolicy, failureCategory, and sanitized errorCode. No stale
  classification score or label is retained.
• Disabled consent: analysisStatus: "not_run", decision, allowed,
  failurePolicy, failureCategory: "consent_required". There is no
  errorCode, classification or billing result. The selected failure policy
  determines allowed; no request is made.
• Every outcome includes an advisory string and
  deliverabilityGuarantee: false.

HTTP 400/413/422 map to malformed, 401/403 to authentication, 402 to
balance, 429 to rate_limit, and other unsuccessful responses to service,
with HTTP_<status> codes. Redirects and network/timeouts map to transport.
Examples of transport codes are REDIRECT_REFUSED, NETWORK_ERROR, TIMEOUT,
FETCH_UNAVAILABLE, and RESPONSE_READ_ERROR. Oversized, non-JSON or invalid
successful responses use RESPONSE_TOO_LARGE, INVALID_JSON, or
INVALID_RESPONSE. Invalid credential strings use INVALID_CREDENTIAL.
Node-side validation uses INVALID_CONFIG, INVALID_INPUT, or
UNKNOWN_FAILURE; these are not spam determinations.

Downstream send nodes must be behind an IF check of
{{ $json.sendrepute.allowed }} equal to Boolean true. Connect the false
branch to explicit review/defer handling. This node does not send or queue mail.
Input JSON remains available downstream and may contain sensitive data; configure
n8n execution-data retention and access controls accordingly.

For contributors, SendRepute.execute is the n8n execution entry point and
SendReputeApi describes credentials. The internal transport helper
classify(apiKey, { sender, subject, body }, { fetch? }) returns a validated
ClassificationResponse or throws SendReputeTransportError with category,
code, and optional HTTP status. Its optional fetch injection exists for
offline testing, not for a workflow-configurable URL or authentication bypass.
Consent and policy enforcement belong to the node; the helper alone does not
enforce them and is not a standalone public SDK API.

Support and security

Use GitHub issues for reproducible, non-sensitive bugs. Account support and
private vulnerability reports: support@sendrepute.com. Never include API keys,
customer messages or unredacted workflow data. See SECURITY.md.
MIT licensed; see LICENSE.