skillZs
★ LIVE SKILL TAGS ★
>>> LIVE SKILLS INDEX <<<
* OPEN SOURCE *
NO LOGIN, NO TRACKING
※ REAL INSTALL DATA ※
← back to all skills
victor-shulga/outbound-engine-skills53 installs

pre-launch-data-check

The last stop-filter before a list goes into the sender. Checks the final file (after enrichment and personalization) across seven layers (source and suppression, address validation, variable completeness, routing, hygiene, signal freshness, activation and capacity) and returns GO, GO WITH FIXES or STOP with the exact rows to fix. Ships a standard-library Python script so the verdict is counted, not guessed. Use when asked to check a list before launch, "is this base ready to send", "перевір базу перед запуском", "можна заливати в Instantly", "pre-launch check", "data integrity scan", "clean the list before upload", or right after waterfall-enrichment and personalization-pipeline and before the first send. Not a fit score (that is prospect-scoring) and not the mailbox check (that is deliverability-audit). This one is about the rows themselves.

How do I install this agent skill?

npx skills add https://github.com/victor-shulga/outbound-engine-skills --skill pre-launch-data-check
view source ↗

Is this agent skill safe to install?

  • Gen Agent Trust Hubpass

    The skill is safe to use and performs data integrity checks on email lists via a local Python script. It has a low security risk associated with indirect prompt injection because it processes untrusted data from user-provided CSV files.

  • Socketpass

    No alerts

  • Snykpass

    Risk: LOW · No issues

What does this agent skill do?

Pre-launch Data Check

Answers one question: can this exact file go into the sender today without hurting the domain, the brand or the test?

It runs on the FINAL file, after waterfall-enrichment and personalization-pipeline, right before upload. Earlier gates decided whether the companies fit (prospect-scoring) and whether the mailboxes can send (deliverability-audit). This one looks at the rows as they will be sent. A fit list with a broken {{first_line}}, a current client in row 212 and 30% catch-all still burns the campaign.

Where it sits

BeforeThis checkAfter
waterfall-enrichment → personalization-pipelinepre-launch-data-check: GO / GO WITH FIXES / STOPdeliverability-audit (mailboxes) → upload

Run it every round, not once per client. Every new batch has its own duplicates, its own stale signals and its own empty fields.

Required tools

Install these before running the skill. It still runs without the ones marked Required, but the result gets noticeably worse: suppression is reported as NOT CHECKED and the validation layer has nothing to read, so the two STOP-level risks go unchecked.

ToolUsed forRequired?Without it
Python 3scripts/check_list.py (standard library only), the counted GO / GO WITH FIXES / STOP verdictRequiredthe seven layers are checked by eye; counts and the verdict are guessed
Icypeas, MoltSets or ZeroBouncethe verification status column the script reads (it does not verify addresses itself)Requiredno verification column means STOP at layer 2
CRM export (Twenty, HubSpot)suppression: everyone already in a conversation, current clientsRecommendedlayer 1 reports suppression as NOT CHECKED
Instantly, HeyReach or lemlistexport of contacts in active campaigns, mailbox count and daily capsRecommendedoverlap with live campaigns is missed; capacity math uses your guess

Inputs the skill expects:

  • The final CSV as it will be uploaded (every column the sequence uses)
  • Suppression files, each as a CSV: CRM export of everyone already in a conversation, contacts in active campaigns, current clients (domains are enough), opt-outs and do-not-contact
  • The list of variables the sequence uses (first_name, company, first_line…)
  • Optional: number of mailboxes and the daily new-contact cap per mailbox

No suppression files = the check still runs, but reports suppression as NOT CHECKED. Say that out loud; the most expensive mistake in outbound is a cold email to a current client.

The seven layers

Each layer is a stop-filter or a fix. Go top to bottom; a STOP in any layer means the file does not go out today.

#LayerWhat is checkedSTOP whenFIX when
1Sourceduplicates inside the file; overlap with CRM, active campaigns, clients, opt-outs; contacts per companyanyone on a suppression listduplicates; more than 3 contacts in one company
2Validationverification status per address; malformed; role addresses; free-mail; email domain vs company domainno verification column at all; any invalid or malformedany risky; catch-all above 25% of the batch; role addresses
3Completenessevery sequence variable present, filled, and rendered (no {{…}}, N/A, null, [NAME])a variable column missing, or empty/unrendered in any rown/a
4Routingevery row has a tier/queue (A/B/D from prospect-scoring) and a sender if you split by sendern/arows with no tier or sender
5Hygienefirst-name case, legal suffixes in company names, empty titlesn/aJOHN, maria, Stripe Inc.: the reader sees these in line one
6Signalssignal age against the freshness window; signal without a source linkn/asignal older than the window (default 60 days); no proof link
7Activationsendable rows after removals vs the volume gate; days of sending at current capacityn/asendable below the volume gate (300 per event-signal hypothesis, 500 per data-point one)

Invalid addresses and suppression hits are STOP, not FIX, for a reason: one bounces the domain, the other emails someone who is already talking to you or paying you. Both cost more than a day's delay.

Run it

python3 scripts/check_list.py final.csv \
  --suppress crm_export.csv active_campaigns.csv clients.csv optouts.csv \
  --vars first_name,company,first_line \
  --mailboxes 6 --daily-cap 30 \
  --out rows_to_fix.csv
  • Columns are auto-detected (email, first_name, company, domain, title, linkedin_url, email_status, tier, signal_date, signal_source). Map anything else with --col email="Work Email".
  • A suppression file with only a domain column suppresses the whole company (clients, DNC accounts).
  • Tune per client: --max-signal-age, --catch-all-max, --max-per-company, --min-volume.
  • Exit code 0 = GO, 1 = GO WITH FIXES, 2 = STOP, so it can sit in a pipeline.
  • --out writes only the rows with problems, with a problems column. Hand that file back to whoever owns the fix.

If the person has no CSV yet (the list lives only in the sending tool), export it first. Do not run the check on a sample; the problems live in the tail.

Process

  1. Ask for the final file and the suppression files. If any suppression source is missing, name it.
  2. Ask which variables the sequence uses. Read them from the sequence if it is at hand.
  3. Run scripts/check_list.py. Read the output; do not re-derive the counts by hand.
  4. For every STOP, say what it costs if ignored and who fixes it.
  5. For FIX items, sort by what the reader would notice first: broken variables and names, then stale signals, then routing.
  6. Give the verdict and one next step. On GO, hand off to deliverability-audit if the mailboxes were not checked this week, otherwise to upload.

Terms and plain-language output

<!-- glossary:start terms=suppression-list,catch-all,role-address,merge-variable,launch-verdict,volume-gate,freshness-window,evidence-url,signal-queue,bounce-rate,enrichment,sequencer -->

Rules for every document a client or their team will read (review, roast, canvas, offer cards, instruction):

  1. Open with «Словник цього документа»: only the terms that actually appear in it, at most 10, each as «термін: що це простими словами». Like the definitions section of a contract.
  2. No internal codes or letter scales in the text (A/B/C, D L T E, BLEND numbers without words, Own/Shared/Generic, Z1.4, GATE 2.5, T1–T6). Write the meaning instead: «є цифра з клієнтом», not «A».
  3. Use the plain form from the table below. English stays only for tool names and for copy that ships in English (promises, hooks, subject lines). If the document is in another language, translate the plain form, never the internal term.
  4. Do not score what cannot exist yet. A block that depends on work not done (signal catalog, case studies, a previous step) gets «пропущено: спершу потрібно <що саме>», not a score.
  5. Every section of a review opens with one concrete action («до наступного дзвінка: …»), then at most five lines of why. No paragraph that only restates the framework.
Internal term (this skill)Write for the readerWhat it means
suppression, suppression list, DNCсписок виключеньпоточні клієнти, ті, з ким уже йде розмова, і ті, хто просив не писати
catch-allдомен, що приймає будь-яку адресуСервер приймає листи на будь-яку адресу, тому перевірити пошту неможливо; пишемо малою партією і стежимо за поверненнями.
role address (info@, sales@)загальна адреса компаніїСкринька відділу без конкретної людини; для персональних листів не годиться, тримаємо окремо.
variable, {{first_line}}, unrenderedзмінна в шаблоні листаполе, яке підставляється в текст; порожнє чи непідставлене видно читачу в першому рядку
GO / GO WITH FIXES / STOPможна запускати · запускати після виправлень · стоппідсумок перевірки бази перед першою відправкою
volume gate, 300 / 500 ruleмінімальний обсяг для тесту300 компаній для гіпотези на події, 500 для гіпотези на факті; менше означає «не перевірено»
freshness window, expiry window, decayтермін свіжості сигналускільки днів сигнал ще дає привід писати; потім він стає просто фактом
signal_evidence_url, proof link, source URLпосилання-доказадреса сторінки, де видно подію або факт; без нього сигнал не рахуємо
Queue A / B / D, event / weak-signal / datapoint queueчерга: свіжа подія · слабший сигнал · без сигналу, лише факткожна черга йде окремою кампанією, щоб середнє не ховало сильний і слабкий сегмент
bounceвідмови доставкичастка листів, які не дійшли, бо адреса неправильна чи мертва.
enrichment, enrichдозбір данихПошук імен, посад і робочих пошт для компаній зі списку з кількох джерел.
sequencer, sending toolсервіс розсилкиінструмент на кшталт Instantly, який відправляє серії за розкладом
<!-- glossary:end -->

Output format

Pre-launch data check: [file], [rows] rows, [date]
Suppression checked against: [files, or NOT CHECKED]

1 Source        [ok / n duplicates / n suppressed]
2 Validation    valid [n%] · catch-all [n%] · risky [n] · invalid [n]
3 Completeness  [each variable: ok / n empty / n unrendered]
4 Routing       [tier split / n unrouted]
5 Hygiene       [n names to fix / n legal suffixes]
6 Signals       [n stale / n without proof / not checked]
7 Activation    sendable [n] vs gate [n] · [n] sending days

VERDICT: [GO / GO WITH FIXES / STOP]
STOP: [numbered, each with the cost of ignoring it]
FIX:  [numbered, most visible first]
Rows to fix: [file]
Next: [one step]

What this check does not do

  • It does not decide fit. A clean row of the wrong company is still the wrong company. That is prospect-scoring and the 90% spot-check in signal-outbound.
  • It does not verify addresses. It reads the verification status your validator wrote; if there is no status, it stops you and sends you back to waterfall-enrichment.
  • It does not test the mailboxes. That is deliverability-audit, and both must be green before the first send.

The idea of checking a base in layers, from source to activation, is adapted from public data-integrity frameworks for revenue teams; the thresholds here are the ones this methodology uses.

Add the canonical catalog link to the repository README so users can inspect current installs and available audits. The publishing guide covers the complete discovery path.

<a href="https://skillzs.dev/skills/victor-shulga/outbound-engine-skills/pre-launch-data-check">View pre-launch-data-check on skillZs</a>