Deliverability Skills for AI Agents

Purpose

Detect when your sending domains or IPs land on public blocklists and produce a delisting action plan before volume drops.

Inputs

• List of sending domains and their mail IPs
• Access to public lookup pages (Spamhaus, Barracuda, SORBS, MXToolbox)
• Recent bounce or block messages if any

Steps

1. For each domain and IP, run lookups against the major public blocklists and record listed or clean status.
2. Capture the listing reason, date, and the specific list name for every hit.
3. Cross reference block bounce messages to confirm which list is actually rejecting mail.
4. Rank each listing by delivery impact, from hard rejects to soft filtering.
5. Pull the exact delisting URL and requirements for each list that flagged you.
6. Draft a remediation note covering send pause, cause, and delisting request steps.

Output

A status table of every domain and IP with listing details, impact rank, and a per-list delisting plan with links.

Purpose

Create a safe daily volume schedule for fresh domains and mailboxes so reputation builds before real campaigns start.

Inputs

• Number of domains and mailboxes
• Domain age in days
• Target daily send volume per mailbox
• Warmup tool in use, for example the pool inside Instantly

Steps

1. Keep every new mailbox on warmup only for the first 14 days, zero cold sends.
2. Start real cold volume at 10 emails per mailbox per day in week three.
3. Increase by roughly 5 emails per mailbox each day until you reach the target.
4. Cap any single mailbox at 30 to 40 cold sends per day regardless of target.
5. Keep warmup running alongside live campaigns at a steady daily count.
6. Pause the ramp for 3 days if reply detection or bounce rate spikes.
7. Spread total volume across mailboxes rather than pushing one hard.

Output

A dated ramp table showing warmup count and cold count per mailbox per day, plus the pause rule and the per mailbox ceiling to respect.

Purpose

Send a controlled test to seed inboxes across major providers so you know whether mail lands in the primary tab, promotions, or spam before a real launch.

Inputs

• A seed list covering Google Workspace, Outlook, and free Gmail and Outlook accounts
• The exact campaign email and subject
• The mailboxes you plan to send from

Steps

1. Send the real campaign copy to the seed list from each production mailbox.
2. Record where each provider placed the message: primary, promotions, or spam.
3. Group results by provider to spot a single failing platform, often Outlook.
4. Check whether spintax variants placed differently than the base copy.
5. If a provider fails, test the same copy from a warmed mailbox to isolate domain versus copy.
6. Re read the failing copy against known spam triggers and links.
7. Repeat the seed test after any fix to confirm placement improved.

Output

A provider by provider placement table, the mailboxes and copy that failed, and a ranked list of the likely causes to fix before scaling volume.

Purpose

Separate genuine negative replies from replies that signal deliverability trouble so you can act on reputation risk fast.

Inputs

• Raw reply export with sender, subject, body, timestamp
• Sending mailbox for each reply
• Baseline reply counts per mailbox

Steps

1. Load the reply export into Google Sheets and normalize columns for body text and mailbox.
2. Pass each body to the LLM and tag it as interested, soft no, hard no, out of office, or complaint.
3. Flag complaint and phrases like stop emailing, reported, and spam as deliverability signals rather than sales signals.
4. Group flagged replies by sending mailbox and count them against total sends per mailbox.
5. Mark any mailbox where flagged replies exceed a set percent threshold for review.
6. Draft a short note per flagged mailbox listing example phrases and suggested pause.

Output

A sheet tab of replies tagged by type plus a per mailbox risk list showing which mailboxes to pause or slow.

Purpose

Watch weekly reply and bounce patterns per sending mailbox so you catch reputation decay before inbox placement falls off.

Inputs

• Export of sends, replies, and bounces per mailbox for the last 4 weeks
• List of active sending domains and their daily send volume
• Current reply rate and bounce rate baseline per domain

Steps

1. Group all sends by domain and mailbox, then compute reply rate and hard bounce rate for each week.
2. Flag any mailbox where reply rate dropped more than 30 percent versus its 4 week average.
3. Flag any mailbox where hard bounce rate crossed 2 percent or auto-reply share spiked.
4. Classify each flagged mailbox as watch, throttle, or pause based on severity.
5. Recommend a send volume adjustment and a re-warm timeline for pause cases.
6. Note which domains share an SPF or reputation pool so risk is not isolated wrongly.

Output

A per-mailbox table with status, trend numbers, recommended volume change, and a short action list for the week.

Purpose

Adjust each mailbox's daily send cap based on its recent reply and bounce behavior so weak inboxes throttle down before they get flagged.

Inputs

• Per-mailbox stats for the last 7 days (sent, replies, bounces)
• Current daily send cap per mailbox
• Minimum and maximum cap limits you allow

Steps

1. Pull sent, reply, and bounce counts per mailbox from the sending platform API.
2. Compute reply rate and bounce rate for each mailbox over the window.
3. Flag mailboxes with bounce rate above 3 percent or reply rate near zero on meaningful volume.
4. Reduce caps by 30 percent for flagged mailboxes, floored at the minimum limit.
5. Increase caps by 10 percent for mailboxes with healthy replies and near zero bounces, capped at the maximum.
6. Hold all others steady and record the reason for each change.
7. Push the new caps back to the platform or export them for manual update.

Output

A table of each mailbox, old cap, new cap, the metric that triggered the change, and a summary of total send capacity shift.

Purpose

Catch mismatched From, Reply-To, and Return-Path domains that quietly hurt trust before a send goes live.

Inputs

• List of sending identities (From address, Reply-To address, sending domain)
• The root brand domain and any approved sub-domains
• Optional: notes on which mailboxes route replies where

Steps

1. Load each identity into a row and split every address into local part and domain.
2. Compare the From domain and Reply-To domain, flag any that do not match or share a root.
3. Check that Reply-To points to a monitored mailbox, not a no-reply or dead address.
4. Confirm the Return-Path domain aligns with the sending domain for DMARC pass.
5. Flag any identity mixing the main corporate domain with a warmup domain.
6. Rank each row as pass, warn, or fail with a one-line reason.
7. Write a short fix note per failing row.

Output

A table of all identities with alignment status, reason codes, and a prioritized fix list ready to hand to the person managing the mailboxes.

Purpose

Verify that a sending domain has correct SPF, DKIM, and DMARC records so mailboxes accept your mail instead of routing it to spam.

Inputs

• The sending domain
• The sending platform, for example Smartlead or Instantly
• Read access to DNS records

Steps

1. Look up the SPF TXT record and confirm it includes your sending platform and ends in a single all mechanism.
2. Check that only one SPF record exists, since two records fail validation.
3. Verify the DKIM selector resolves and the public key is present, using MXToolbox.
4. Confirm a DMARC record exists at the correct host with a policy of none, quarantine, or reject.
5. Check that DMARC alignment matches the From domain used in campaigns.
6. Test one live send and read the received headers for pass results on all three.
7. List any record that is missing, duplicated, or misaligned.

Output

A pass or fail line for SPF, DKIM, and DMARC, the exact record values found, and the corrected record text to paste into DNS for anything failing.

Purpose

Turn soft bounce events into a controlled retry schedule so temporary failures do not burn sender reputation.

Inputs

• Export of bounce events with recipient, timestamp, and SMTP response code
• Current daily send caps per mailbox
• Max retry attempts you allow per contact

Steps

1. Pull bounce rows and separate hard bounces (5xx) from soft bounces (4xx) using the response code.
2. Group soft bounces by reason: mailbox full, greylisting, rate limit, or temporary server error.
3. Assign a backoff interval per reason, for example 6 hours for greylisting and 48 hours for mailbox full.
4. Cap total retries per contact and flag anything past the cap for suppression.
5. Spread scheduled retries across mailboxes so no single sender exceeds its daily cap.
6. Write a retry queue with contact, next attempt time, and assigned mailbox back to the sheet.

Output

A retry schedule table plus a suppression list for exhausted contacts, ready to load into your sending tool.

Purpose

Isolate the source of a spam trap hit and remove the risk before reputation drops further.

Inputs

• Bounce and complaint export from your ESP
• List of recent send campaigns with source tags
• Contact acquisition dates and enrichment provider per row

Steps

1. Pull all hard bounces and complaint events flagged as trap or unknown user for the last 30 days.
2. Join each flagged address back to its acquisition source, enrichment provider, and import batch.
3. Group hits by source and calculate a hit rate per source to find the worst offender.
4. Check whether flagged addresses are role accounts, old imports, or catch-all domains that slipped through verification.
5. Quarantine every address tied to the highest hit rate source into a suppression segment.
6. Draft a corrective rule, such as blocking that provider or adding a re-verification step before import.

Output

A ranked table of trap hits by source, a suppression list of flagged addresses, and one written rule to prevent recurrence.

Purpose

Review a draft email for language and formatting that commonly trips spam filters, then return a cleaner version.

Inputs

• The email body and subject line
• The sending domain
• Any links or images in the email

Steps

1. Flag classic trigger words such as free, guarantee, act now, and risk free.
2. Flag heavy punctuation, all caps words, and more than one exclamation mark.
3. Count links and recommend at most one, never a bare tracking link in touch one.
4. Check that the link domain matches or relates to the sending domain.
5. Flag image only sections and any large signature block with logos.
6. Check the text is under 120 words and free of hidden HTML styling.
7. Rewrite each flagged item with a neutral alternative that keeps the meaning.

Output

A list of flagged items with severity, a clean rewritten email body and subject, and a short note on which change matters most for placement.