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-checkIs 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
| Before | This check | After |
|---|---|---|
waterfall-enrichment → personalization-pipeline | pre-launch-data-check: GO / GO WITH FIXES / STOP | deliverability-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.
| Tool | Used for | Required? | Without it |
|---|---|---|---|
| Python 3 | scripts/check_list.py (standard library only), the counted GO / GO WITH FIXES / STOP verdict | Required | the seven layers are checked by eye; counts and the verdict are guessed |
| Icypeas, MoltSets or ZeroBounce | the verification status column the script reads (it does not verify addresses itself) | Required | no verification column means STOP at layer 2 |
| CRM export (Twenty, HubSpot) | suppression: everyone already in a conversation, current clients | Recommended | layer 1 reports suppression as NOT CHECKED |
| Instantly, HeyReach or lemlist | export of contacts in active campaigns, mailbox count and daily caps | Recommended | overlap 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.
| # | Layer | What is checked | STOP when | FIX when |
|---|---|---|---|---|
| 1 | Source | duplicates inside the file; overlap with CRM, active campaigns, clients, opt-outs; contacts per company | anyone on a suppression list | duplicates; more than 3 contacts in one company |
| 2 | Validation | verification status per address; malformed; role addresses; free-mail; email domain vs company domain | no verification column at all; any invalid or malformed | any risky; catch-all above 25% of the batch; role addresses |
| 3 | Completeness | every sequence variable present, filled, and rendered (no {{…}}, N/A, null, [NAME]) | a variable column missing, or empty/unrendered in any row | n/a |
| 4 | Routing | every row has a tier/queue (A/B/D from prospect-scoring) and a sender if you split by sender | n/a | rows with no tier or sender |
| 5 | Hygiene | first-name case, legal suffixes in company names, empty titles | n/a | JOHN, maria, Stripe Inc.: the reader sees these in line one |
| 6 | Signals | signal age against the freshness window; signal without a source link | n/a | signal older than the window (default 60 days); no proof link |
| 7 | Activation | sendable rows after removals vs the volume gate; days of sending at current capacity | n/a | sendable 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
domaincolumn 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.
--outwrites only the rows with problems, with aproblemscolumn. 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
- Ask for the final file and the suppression files. If any suppression source is missing, name it.
- Ask which variables the sequence uses. Read them from the sequence if it is at hand.
- Run
scripts/check_list.py. Read the output; do not re-derive the counts by hand. - For every STOP, say what it costs if ignored and who fixes it.
- For FIX items, sort by what the reader would notice first: broken variables and names, then stale signals, then routing.
- Give the verdict and one next step. On GO, hand off to
deliverability-auditif 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):
- 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.
- 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».
- 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.
- 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.
- 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 reader | What it means |
|---|---|---|
| suppression, suppression list, DNC | список виключень | поточні клієнти, ті, з ким уже йде розмова, і ті, хто просив не писати |
| catch-all | домен, що приймає будь-яку адресу | Сервер приймає листи на будь-яку адресу, тому перевірити пошту неможливо; пишемо малою партією і стежимо за поверненнями. |
| role address (info@, sales@) | загальна адреса компанії | Скринька відділу без конкретної людини; для персональних листів не годиться, тримаємо окремо. |
| variable, {{first_line}}, unrendered | змінна в шаблоні листа | поле, яке підставляється в текст; порожнє чи непідставлене видно читачу в першому рядку |
| GO / GO WITH FIXES / STOP | можна запускати · запускати після виправлень · стоп | підсумок перевірки бази перед першою відправкою |
| volume gate, 300 / 500 rule | мінімальний обсяг для тесту | 300 компаній для гіпотези на події, 500 для гіпотези на факті; менше означає «не перевірено» |
| freshness window, expiry window, decay | термін свіжості сигналу | скільки днів сигнал ще дає привід писати; потім він стає просто фактом |
| signal_evidence_url, proof link, source URL | посилання-доказ | адреса сторінки, де видно подію або факт; без нього сигнал не рахуємо |
| Queue A / B / D, event / weak-signal / datapoint queue | черга: свіжа подія · слабший сигнал · без сигналу, лише факт | кожна черга йде окремою кампанією, щоб середнє не ховало сильний і слабкий сегмент |
| bounce | відмови доставки | частка листів, які не дійшли, бо адреса неправильна чи мертва. |
| enrichment, enrich | дозбір даних | Пошук імен, посад і робочих пошт для компаній зі списку з кількох джерел. |
| sequencer, sending tool | сервіс розсилки | інструмент на кшталт Instantly, який відправляє серії за розкладом |
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-scoringand the 90% spot-check insignal-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.
How can the creator link this skill?
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>