Terminal
Terminal/Skill
Checks whether email addresses will accept mail before you send, returning a risk tier for each address.
By Clay
Works with
npx skills add clay-run/clay-skill-creator --skill verify-email-deliverabilityThe insight: "deliverable" is a tier, not a boolean — and every validator hides the
riskiest tier behind its safest-looking field, in opposite directions. Verified live:
ZeroBounce returns catch-all as status: valid (risk in sub_status); Enrichley returns
it as valid: false (signal in result). Reading one top field either blesses bounces or
discards live mailboxes — the verdict is always the field pair. The reverse trap: a live,
monitored support@ comes back do_not_mail — validators grade for cold-list hygiene,
not readership.
Nothing here ships with a value. Each one is the installer's, not the author's: ask for it, never substitute a plausible default, and where an answer does not exist say which step becomes unavailable rather than guessing. Where a default IS defensible it is named below, and using it means saying so in the output.
| Input | What the installer supplies | If it is missing |
|---|---|---|
| The addresses | exact strings, one per row | no default, and never construct or repair an address — a corrected address is a different address |
| What they are for | a one-off reply, or a volume send | ask. It changes which confidence tiers are usable at all, not just the reporting |
If an answer sheet is present beside this skill, load it and ask only for what it does not cover. A partial sheet is normal; a value it is missing gets asked for on its own rather than restarting the interview. Say which values came from the sheet before using them — a sheet applied silently is a wrong field nobody catches. If there is no sheet, say nothing about sheets — the check is a file lookup, not a question, so run the interview as though the feature did not exist rather than reporting an absence. At delivery, offer to save the answers back (identifiers only — never a token or a password), private and never published — and phrase the offer so it explains itself: "want me to save your answers to a file, so the next person on your team doesn't have to answer these again?"
Run clay whoami; echo "exit_code=$?". If it fails or Clay tools are missing, run the
Clay plugin's setup skill (or follow
https://raw.githubusercontent.com/clay-run/agent-plugins/main/GETTING_STARTED.md), restart
if told to, and re-run this skill. Tell the user which workspace you're in.
Before paying, resolve MX for each distinct domain (any DNS tool). Three shapes:
NXDOMAIN / no MX → guaranteed bounce — report invalid free, skip the paid check.
Null MX (single record 0 .) → the domain declares it takes no mail — invalid, even
though a record "exists". Real MX → proceed.
Validators are catalog actions, not a managed function — workspaces differ. Discover with
clay workflows actions list; fetch the input schema with clay workflows actions schema — never guess field names. Prefer the lowest priorityTier (tier 2 =
Clay-managed account, typically ZeroBounce or Enrichley at ~0.1 credits/check; tier 4 =
user's own key — never ask for a key in chat). Vocabulary matters as much as price: you
need a validator that separates catch-all from valid — a binary provider can't produce
the tiers. State per-check cost before running; report the actual charge from run usage
metadata (billing differs even within a tier).
For 1–20 addresses, run the action directly (Clay MCP execute_clay_action; input
{"email": "..."}). Leave "only safe to send" switches OFF — they collapse the tiers
you're here to report. Checks return in seconds, every verdict. For hundreds, recurring
cleans, or addresses you're finding rather than checking, hand off to
clean-email-list or find-work-email and say so.
Catch-all isn't a dead end. If the send matters, run Enrichley on the same address: it
probes the specific mailbox — result: catch_all_validated means confirmed live (treat
as valid, verified live); plain catch_all stays unconfirmable. Production builds send
to the former, throttle or skip the latter.
Map the observed fields to the verdict:
valid + sub_status: "" (empty string, not null), or Enrichley
result: ok / catch_all_validated → valid.valid + sub_status: catch_all, or Enrichley result: catch_all →
catch-all (risky) — never report as plain valid; offer Step 5.do_not_mail (+ role_based, role_based_catch_all, disposable,
global_suppression) → flagged — may be genuinely read (support@!) but poison for
volume; report the sub-reason, the user decides per use.invalid (+ mailbox_not_found, does_not_accept_mail) → invalid.unknown, timeouts, empty payloads → could not verify — never round either way,
and never gate on run status: it reports SUCCESS for every verdict; only payload fields
are data.Per address: email · verdict (valid / catch-all risky / catch-all validated / flagged: reason / invalid / could not verify) · raw fields · provider. Multi-address runs add %
per tier plus a one-line recommendation for the stated use.
Ask: "Can I send to jordan.lee@northfield.example and hello@initech-consulting.example?"
MX pre-check: both domains have real MX — proceed. ZeroBounce (~0.1 credits each,
approved): first returns valid / sub_status: "" → valid. Second returns valid /
sub_status: catch_all; the send matters, so Enrichley on the same address returns
catch_all_validated → valid (probe-confirmed). Summary: 2 sendable, ~0.3 credits.