Cleared To Call

Cleared to Call

Use this skill when an agent is about to place an outbound call to a consumer about money: a past-due balance, a payment reminder, a collection follow-up, or any account-servicing call where the recipient did not initiate the contact.

cleared-to-call supplies offline advisory helpers, not a dialer or a legal compliance certification. They read supplied records, evaluate a declared policy, and print results. They do not place or end calls, write suppression entries, or create or verify an audit chain. The integration requirements below describe work a host must implement; they are not shipped runtime features. An ALLOW result is not proof that a call is lawful or authorized.

The rules are data, not prose. They live in assets/policy.json and are explained in references/policy.md.

When To Use

Use this skill for:

• outbound collection, payment-reminder, or past-due account calls
• any batch of consumer accounts where each row must be checked before dialing
• workflows that must honor "stop calling me" during the call, not after it
• workflows collecting evidence for a human compliance assessment
• workflows explaining why the declared policy blocked a record

When Not To Use

Do not use this skill to:

• place calls outside the United States, or under any rule set other than US federal TCPA, FDCPA, and Regulation F - the policy is single-jurisdiction and a different region needs a different policy file, not a workaround
• decide anything for inbound calls the consumer initiated
• override, soften, or "just this once" bypass a block reason
• guess a missing timezone, consent record, or phone number
• give the recipient financial, legal, credit, or settlement advice
• negotiate, threaten, or discuss consequences of non-payment

If the caller wants a block overridden, the answer is to fix the underlying record, not the gate. A missing timezone is fixed by getting the timezone, not by assuming one.

Required Inputs

Each account must carry all of these. None of them may be inferred:

• account_id
• display_name
• phone_e164 - E.164, for example +15550101234
• timezone - IANA zone name, for example America/New_York
• amount_due and currency
• consent_on_file - true or false
• consent_timestamp - required when consent_on_file is true

Check a batch before evaluating any of it:

node scripts/validate-input.mjs --file accounts.json

A row that fails validation is reported and skipped. It is never dialed.

Never derive timezone from the phone number, the area code, the operator's locale, the server clock, or the country code. Rule 1 is a statement about the recipient's local time, so a guessed timezone is a guessed legal position. If the timezone is missing or unknown, the account is blocked.

Try It

The skill ships a fictional batch that hits every branch. Nothing dials:

node scripts/validate-input.mjs --file assets/example-accounts.json
node scripts/evaluate-account.mjs \
  --file assets/example-accounts.json --account-id A-1002 \
  --suppression assets/example-suppression.jsonl --now 2026-08-28T13:30:00Z
node scripts/check-revocation.mjs --utterance "stop calling me"

At that instant A-1001 is clear, A-1002 is outside its local call window, A-1003 has no consent, and A-1004 is suppressed.

The Pre-Dial Gate

The host must run these four checks before dialing. Passing them is necessary for this sample policy, not sufficient authorization for a real call.

Run the gate on one account:

node scripts/evaluate-account.mjs \
  --file accounts.json \
  --account-id A-1001 \
  --suppression suppression.jsonl

Exit code 0 is ALLOW, 2 is BLOCK, 1 is unusable input. The printed JSON holds every rule result, not only the failing one, so an audit reader can see the whole evaluation.

When several rules fail, the reported block_reason is the first failure in policy order. All four are still evaluated and printed for the host to retain.

Read references/policy.md for what each rule means, which authority it comes from, and the property it is meant to enforce.

The In-Call Revocation Handler

Rule 5 cannot be decided before dialing. A host integration has two duties; only the literal transcript checker is included here:

1. The call task tells the agent to stop the moment the recipient revokes: acknowledge once, end the call, do not ask why, do not offer alternatives.
2. After the call, run the transcript checker, then have the host persist any required suppression entry and record the outcome.

The checker prints a required action; it does not carry it out. If the agent talks past an opt-out, the host must still persist suppression and record opt_out. A post-call check cannot retrospectively stop a running call.

node scripts/check-revocation.mjs --utterance "stop calling me"
node scripts/check-revocation.mjs --transcript transcript.json

Exit code 3 means the number must be suppressed. Only recipient turns count: the agent saying "if you would like us to stop calling" is a required disclosure, not an opt-out.

Once the host persists suppression and supplies the updated list, R3 blocks the number on later evaluations. The host must retain the opt-out unless the consumer asks for it to be reversed.