check-deliverability
Skill1 starsChanged 5 months agoMITView on GitHub
Diagnose AND fix cold email deliverability incidents. Use when the user mentions "deliverability," "inbox health," "warmup status," "bounce rate," "inbox placement," "domain health," "infra monitoring," "reply rate dropped," "campaign died," "domain blacklisted," "warmup blocked," "stuck in Promotions," "in spam," or "deliverability fire." Runs Phase 1 (diagnose) → if anything red, Phase 2 (fix runbooks).
- Reads credentials
1 other project uses a version of this.
Adopted by
Most recently taken up by
- MattSezgin/unofficial-smartlead-clitheir own version, first seen May 23, 2026
Other versions1 project uses a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 1
- MattSezgin/unofficial-smartlead-cli★ 8same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
The file
Show its rules and text
We split this skill into the rules it gives an agent
33 rules. About 0 are common, 30 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈30
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 14
Run these in addition to a basic health check.
- Very specific: line 17
DNS auth audit: Run SPF/DKIM/DMARC checks (see references/diagnostics.md) against every active sending domain on demand.
- Distinctive: line 34
Before launching a new campaign (pre-flight check)
- Distinctive: line 60
Run the quick health check first
- Distinctive: line 61
Look at the per-domain breakdown (group inboxes by domain)
- Distinctive: line 62
Check for the Outlook/Azure decay pattern:
- Distinctive: line 64
If detected, recommend domain rotation or a warmup increase
- Distinctive: line 65
Check for provider-specific issues (all Google down vs all Outlook down)
- Distinctive: line 66
Look at campaign-level data to isolate whether the issue is infrastructure or copy
- Distinctive: line 102
Always check infrastructure health BEFORE suggesting campaign changes
- Distinctive: line 103
If bounce rate > 2%, investigate infrastructure first (don't blame copy)
- Distinctive: line 104
If reply rate drops but bounce rate is fine, then investigate copy/targeting
- Distinctive: line 106
Never pause warmup on healthy inboxes - warmup runs alongside campaigns
- Distinctive: line 112
If Phase 1 surfaced anything yellow or red, identify the symptom and run the matching runbook.
- Distinctive: line 128
Before invoking Runbook A on a reply drop, classify the shape of the drop:
- Distinctive: line 153
Bounce check: - if >3%, run Runbook B in parallel.
- Distinctive: line 168
Test placement on existing infra if it's not too far gone.
- Distinctive: line 205
Run your domain-rotation process.
- Distinctive: line 209
If bounces concentrate at specific recipient providers, MX-lookup the bouncing domains.
- Distinctive: line 232
Default = flag-and-confirm.
- Distinctive: line 292
DNS auth verify: all sending domains in the affected cohort.
- Distinctive: line 297
If DNS auth fails → fix DNS, re-test, wait 7d.
- Distinctive: line 298
If domain reputation = BAD → run Runbook G (the 550 5.7.708 path) - a Microsoft support ticket is likely needed.
- Distinctive: line 299
If routing was wrong → segment the list, route Microsoft-recipient leads to a Microsoft 365 cohort only, kill the cross-routed batch.
- Distinctive: line 300
If all the above is clean → strip the Outlook cohort from the active campaign, age the domains another 30d, swap to fresh Microsoft 365 …
- Distinctive: line 319
Long-term: consider migrating to a dedicated-IP tier (dedicated > shared at any premium worth paying once this happens).
- Distinctive: line 328
When an Outlook/Proofpoint cohort underperforms and you need to keep the campaign sending while you fix:
- Distinctive: line 345
Flag and confirm with a human: - never auto-execute (domain-wide blast radius too high).
- Distinctive: line 365
When multiple runbooks trigger at once or an account is at ~0 reply rate:
- Very specific: line 370
DNS auth verify: all sending domains (references/diagnostics.md).
- Distinctive: line 390
Wait 7 d post-restart. Re-run Phase 1 to confirm recovery.
- Distinctive: line 392
If domains/inboxes were replaced, expect 2 weeks of warmup before full recovery.
- Distinctive: line 393
Document: cause, fix, time-to-recovery.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Deliverability Monitor + Incident Response
Diagnose health AND run the fix runbook in one flow. Whenever campaigns underperform - or on a regular cadence - invoke this skill. It runs Phase 1 (audit) and, if anything trips a threshold, walks through Phase 2 (the matching runbook).
This skill is self-contained: every threshold and runbook lives here. It's platform-agnostic in language - examples lean on the common cold-email sending platforms (Instantly, PlusVibe, Smartlead, EmailBison), but the method is the same wherever you send. Swap the tool calls for whatever platform you run.
Phase 1 - gate audits
Run these in addition to a basic health check. Findings format: {symptom, severity, affected_domains, recommended_runbook}. Severity ≥ yellow on any of (reply-drop / bounce / blacklist / warmup-blocked / placement) drops directly into the matching runbook in Phase 2.
- Domain-level (not just inbox-level) reputation check: Group inboxes by domain. If ANY inbox on a domain shows spam-fold or warmup-blocked, flag the entire domain as a zombie candidate. Recommend a flag-and-confirm quarantine: kill all mailboxes on the domain (cancel the subscription first, then delete from the sequencer). Never auto-execute - the blast radius is too high.
- DNS auth audit: Run SPF/DKIM/DMARC checks (see
references/diagnostics.md) against every active sending domain on demand. Any miss = surface it in the alert. - The 1% rule: Auto-flag any inbox or domain with ≥200 sent and <1% reply rate for triage/retirement. Catches dead-from-day-one inboxes that the week-over-week drop trigger misses.
- Bounce thresholds, tiered: <1% green / 1-2% normal / 2-3% yellow (verify list) / 3-5% red (pause) / >5% stop immediately. Use this tiered table in alerts, not a flat 2%/3% line.
- Spam placement thresholds: ≥90% inbox great / 80-90% acceptable / 70-80% yellow / <70% red. Surface from placement-test data when available.
- Warmup-blocked flag: Surface the platform's warmup-blocked / account-suspended state. Treat it as a separate failure mode from spam-fold - warmup-network rejection is a domain-level zombie signal.
- Inbox reputation <98% (warmup score): flag for review. <90% or warmup-blocked = auto-kill candidate.
- Domain bottom-decile ranking: per sending account, rank domains by an inbox-reputation proxy. The bottom 10% are kill candidates.
- Microsoft shadow-ban diagnostic: 0 bounces + 0 replies on ≥200 sends to a Microsoft-recipient cohort = Outlook silent-block suspect. Routes to Runbook F.
- Spam-complaint thresholds: >0.1% target / >0.3% hard cap = pause (pull from your Postmaster/feedback-loop data when a domain is enrolled).
- Variant fatigue: any single copy variant past ~75K sends in 30 days = retirement candidate. Complements the 25% week-over-week relative trigger.
- 48h open-tracking diagnostic (last-resort burn test): when the signal checks are inconclusive on a suspected burn, toggle open-tracking on for 48h. A healthy mailbox shows ~60% opens (Apple Mail auto-fetch baseline); a burned one shows ~10%. Toggle it back off after - never leave open-tracking on past the diagnostic (it triggers Gmail's Promotions tab).
When to Use
- User asks about deliverability, inbox health, warmup status
- User wants to check bounce rates or inbox placement
- User mentions domain health or infra monitoring
- Before launching a new campaign (pre-flight check)
- Investigating why reply rates dropped
Data Sources
Pull from the platform you actually send on. Most platforms expose either an MCP server or a REST API with the three things you need: a list of email accounts (with status), warmup analytics (score + placement), and campaign analytics (sent / replies / bounces / opens).
- Via MCP (e.g. an Instantly, Smartlead, or PlusVibe MCP server): list accounts, get warmup analytics, get campaign analytics.
- Via REST (example, Instantly v2):
GET /api/v2/campaigns/analytics/steps?campaign_id=CAMPAIGN_ID&include_opportunities_count=truewithAuthorization: Bearer {YOUR_API_KEY}.
Store your API key in an env file / secrets manager - never hard-code it. Concrete commands for the common platforms are in references/diagnostics.md.
Optional - automate it: a scheduled job (cron / a serverless function) that snapshots these metrics a few times a day, stores them as JSONL for trend analysis, and pings you (Slack/Telegram/email) when a threshold trips. The audits above are the logic; the schedule just runs them while you sleep.
Workflow
Quick Health Check
- Pull all email accounts
- Count: total, active, disconnected, suspended
- Pull warmup analytics
- Flag any inbox with warmup score < 90 or status != active
- Pull campaign analytics for active campaigns
- Compute: reply rate, bounce rate, open rate
- Compare against the thresholds below
Deep Investigation
- Run the quick health check first
- Look at the per-domain breakdown (group inboxes by domain)
- Check for the Outlook/Azure decay pattern:
- Good results for 3-4 weeks, then a sudden drop
- If detected, recommend domain rotation or a warmup increase
- Check for provider-specific issues (all Google down vs all Outlook down)
- Look at campaign-level data to isolate whether the issue is infrastructure or copy
Thresholds
| Metric | Target | Warning | Critical |
| Warmup score | > 95 | < 90 | < 80 |
| Bounce rate (inbox) | < 1% | > 3% | > 5% |
| Bounce rate (domain) | < 1% | > 2% | > 3% |
| Reply rate (campaign) | > 1.5% | < 0.5% | < 0.2% |
| Disconnected inboxes | 0 | > 0 | > 3 per domain |
These are starting points. Calibrate to your own book - some verticals structurally run looser, and a number that's "below average" in one can be the right call in another given the ACV. Data, not feelings.
Output Format
Present results as a summary table, then flag any issues:
DELIVERABILITY CHECK - [date]
Accounts: X total, Y active, Z disconnected
Campaigns: X active, Y emails sent today
ALERTS (if any):
- [CRITICAL] domain.com: 2/3 inboxes disconnected
- [WARNING] trydomain.com: warmup score 84.2 (below 90)
DOMAIN HEALTH:
| Domain | Provider | Inboxes | Warmup | Status |
|--------|----------|---------|--------|--------|
| ... | google | 3/3 | 96.1 | OK |Rules
- Always check infrastructure health BEFORE suggesting campaign changes
- If bounce rate > 2%, investigate infrastructure first (don't blame copy)
- If reply rate drops but bounce rate is fine, then investigate copy/targeting
- For Outlook domains, check for the 4-5 week decay cycle before other causes
- Never pause warmup on healthy inboxes - warmup runs alongside campaigns
Phase 2: Fix Runbooks
If Phase 1 surfaced anything yellow or red, identify the symptom and run the matching runbook. Multiple symptoms at once → skip to the 72-hour total-failure checklist.
Symptom selector
| Symptom | Fast tell | Runbook |
| Reply rate dropped sharply | >25% week-over-week drop OR <1% on ≥200 sends | A |
| Bounce rate spiked | >3% bounces this week | B |
| Domain blacklisted | MXToolbox / Spamhaus / SURBL hit | C |
| Warmup blocked / mailbox flagged | Platform surfaces "warmup blocked" or equivalent | D |
| Gmail Promotions tab | Test send to a fresh @gmail.com lands in Promotions, not Primary or Spam | E |
| Outlook shadow-ban suspect | 0 bounces + 0 replies on ≥200 sends to a Microsoft cohort | F |
| 550 5.7.708 in bounce text | Microsoft-source bounces with this code | G |
Reply-drop pattern classifier
Before invoking Runbook A on a reply drop, classify the shape of the drop:
- Overnight 90% reply collapse (single day, sharp): copy-fingerprint burn. Run A.5 (build new copy before new infra) FIRST. Most likely a single phrase that mailbox providers started fingerprinting.
- Gradual decline over weeks: domain burn. Run A.4 (placement-test + DNS auth + bounce check) FIRST. Most likely cumulative mailbox-pool degradation.
Different shapes route to different entries within Runbook A.
Full symptom→cause→fix matrix → references/symptom-matrix.md.
Diagnostic curl/dig commands → references/diagnostics.md.
Runbook A - "reply rate dropped"
A.1 Quantify the drop
- Pull the last 4 weeks' reply rate per cohort.
- Real drop = current week <50% of the prior 4-week average, OR current rate <1% on ≥200 sends (the 1% rule).
- Below threshold → noise, wait one more week. Above → continue.
A.2 Fleet-wide or campaign-specific?
Pull all active campaigns. ALL dropping → fleet-level (infra or copy fingerprint). One → campaign-specific (targeting / copy / list).
A.3 Fleet-wide drop - check in this order
- DNS auth check -
digSPF/DKIM/DMARC perreferences/diagnostics.md. Any miss → fix at the registrar, re-test, continue. - Lightweight placement test - duplicate the sequence, send to your own seed inboxes (1 Google Workspace + 1 Microsoft 365). Spam on the Workspace seed = a real deliverability problem (continue to A.4). Inbox on both = look at copy/targeting.
- Warmup health check - per inbox. Flag reputation <50, blocked status, or score drops.
- Bounce check - if >3%, run Runbook B in parallel.
- Diagnose copy first. Reply rate dropped while ICP/volume/infra are unchanged → copy IS the cause. Skip to A.5.
A.4 Spam-placement problem - time-ordered fix
| When | Action |
| Today | Pause the highest-volume affected campaign |
| Day 1 | Lightweight placement test on 2 sender subsets (active vs warming pool). Find the worst cohort. |
| Day 1 | DNS auth + mxtoolbox.com/blacklists.aspx per domain |
| Day 2 | Fix invalid DKIM, simplify HTML, strip Email-1 links (the hard rule = zero links in Email 1) |
| Day 3-14 | Cut daily send/inbox 50% for 1 week |
| Day 14 | Re-test placement. ≥85% inbox → resume. <70% → escalate to replace-vs-repair (Runbook C) |
A.5 Build new copy before new infra
- 5+ structural variants. Spintax on each. A 20+ subject-line pool.
- Test placement on existing infra if it's not too far gone.
- Any new infra batch ships with new copy on day one. Burned copy never touches new mailboxes (hard gate - it just re-burns them).
Runbook B - "bounce rate spiked >3%"
B.1 What kind of bounces?
Hard (invalid address) vs soft (greylist / temporary). Apply the tiered table:
| Bounce % | Action |
| 1-2% | Normal; monitor |
| 2-3% | Yellow; verify list before next send |
| 3-5% | Red; pause campaign |
| >5% | Stop immediately to avoid ISP suspension |
B.1.5 Reply-rate × bounce-rate combined gate
The default "spintax aggressively before pausing on bounce alone - copy burns before domains" doctrine is correct ONLY when the campaign is otherwise healthy. Combine with reply rate to choose the path:
| Reply rate (per domain) | Bounce rate | Action |
| ≥1% | 3-5% (red) | Spintax-first. Aggressive spintax on copy, monitor 7d. Campaign keeps running. |
| ≥1% | >5% | Pause. Investigate list quality (B.2) before resuming. |
| <1% (the 1% rule) | 3-5% (red) | Auto-swap the domain. Healthy domains don't co-occur with both signals - this combination = the domain is structurally burned, no copy iteration recovers it. Run the Runbook C zombie protocol on it. |
| <1% | <3% | Runbook A (reply-rate problem, not bounce). |
Why this combined gate: "spintax before pause" holds when copy is the variable AND domain reputation is intact (reply rate proves it). When BOTH signals trip together, the domain has structurally degraded - spintaxing copy on a burned domain just delays the rotation, costing more sends per recovered positive than swapping cohorts immediately.
B.2 List problem (mostly hard)
- Export the remaining leads.
- Re-verify via a validation API (e.g. a real-time email verifier). Reuse only
status=ok. - Re-check
unknown/inbox_fullresults after 7 days (some resolve to valid). - Reload the cleaned list. Stamp your DB with
verified_at+last_contactedso you don't re-pitch contacted leads.
B.3 Reputation problem (mostly soft)
- Cut daily volume per inbox 50% for 2 weeks.
- Run your domain-rotation process.
- Confirm warmup is at a 1:1 ratio with cold sends.
B.4 MX gateway mismatch
If bounces concentrate at specific recipient providers, MX-lookup the bouncing domains. Barracuda / Mimecast / Proofpoint / MXLogic dominant → you're sending from Google inboxes into enterprise gateways. Segment the list, route enterprise leads through a Microsoft 365 / SMTP cohort only, in a separate campaign.
Runbook C - "domain blacklisted"
C.1 Confirm
mxtoolbox.com/blacklists.aspx. Spamhaus DBL, SURBL, Barracuda. See references/diagnostics.md.
C.2 Replace vs repair
| Domain age | Status | Action |
| <30 d | Blacklisted | Replace. Archive, buy a lookalike, warm 2 weeks before reuse. |
| 30-90 d | Blacklisted | Judgment. Lean replace if it's a fresh blacklist; lean repair if it was previously good. |
| >90 d | Blacklisted (was good) | Repair. Stop sending 7 d, file a delisting request, slow-resume at 25% volume. |
C.3 Zombie protocol
One mailbox blacklisted on a domain → ALL mailboxes on that domain inherit the problem:
- Lightweight placement test from one mailbox per affected domain.
- Domain lands in spam → flag and confirm with a human → kill ALL mailboxes on that domain.
- Cancellation order matters: cancel the subscription at the provider FIRST, delete from the sequencer SECOND. Reverse order = zombie billing (you pay for mailboxes that no longer send).
- Replace from the warming pool.
Default = flag-and-confirm. Never auto-execute domain-wide kills.
C.4 IP-pool issue
Reply rates drop simultaneously across ALL accounts on the same provider → likely shared-IP contamination. Contact the provider; if it's shared, wait 1-2 weeks for rotation OR migrate to a dedicated IP.
Runbook D - "warmup blocked / mailbox flagged"
D.1 Identify
Your platform surfaces a warmup score that's dropping, or a "blocked"/"suspended"/health=BAD state. Treat any of those as the trigger.
D.2 Triage by inbox age
| Age | Status | Action |
| <30 d | Burned | Retire. Replace from the warming pool. Disable warmup, cancel sub, delete from sequencer (in that order). |
| 30-90 d | Flagged | Judgment. Strong history = repair. Otherwise retire. |
| >90 d | Was good | Repair. Disable warmup, wait 48h, re-enable at ~15 warmup emails/day (vs the usual 40). Monitor 14 d. |
D.3 Domain-level check
Multiple inboxes on the same domain flagged → run the Runbook C zombie protocol.
D.4 Cause check
Common cause: the warmup emails share a copy fingerprint with active campaigns. Confirm your warmup config doesn't reuse active-campaign copy. An insurance/warming pool runs at lower volume (~15/day, no daily ramp).
Runbook E - "Gmail Promotions tab"
E.1 Test on a fresh Gmail you control
Send Email 1 to a fresh @gmail.com (not in the campaign):
- Primary → not the issue.
- Promotions → this runbook.
- Spam → run Runbook A.
E.2 Triggers + fixes
Most things that push Email 1 into Promotions are formatting violations:
| Trigger | Fix |
| ≥1 link in Email 1 | Strip ALL links (hard rule) |
| Inline images | Remove |
| Heavy HTML / inline CSS | Simplify to near-plain-text |
| "click here", "act now", "limited time" | Banned-words scan against your spam-word list |
| Hyperlinked company name in signature | Strip the .com |
| Logo / social icons in signature | Text-only signature |
| Tracking pixel | Disable open tracking platform-wide |
E.3 A/B
Current vs stripped. 50 leads each. 7-day reply rate. Promote the winner.
Runbook F - "Microsoft shadow-ban suspect"
F.1 Identify
Symptom: ≥200 sends to a Microsoft-recipient cohort in 14d with 0 bounces AND 0 human replies. Microsoft is silently blocking without telling you (no bounce, no spam-fold flag, just a black hole).
F.2 Confirm before acting
- Lightweight placement test to a fresh
@outlook.comseed inbox. Is the mail landing in spam, or vanishing entirely? - DNS auth verify all sending domains in the affected cohort. Microsoft is stricter than Google on auth.
- Check Postmaster/feedback data for any enrolled domain in the cohort - if reputation is BAD/LOW, Microsoft has tagged you and the block is reputation-driven, not transport.
- MX-gateway segmentation check: are Google inboxes incorrectly sending to Microsoft-gateway recipients? If yes, fix routing first - shadow-bans are often just wrong-cohort routing.
F.3 Fix
- If DNS auth fails → fix DNS, re-test, wait 7d.
- If domain reputation = BAD → run Runbook G (the 550 5.7.708 path) - a Microsoft support ticket is likely needed.
- If routing was wrong → segment the list, route Microsoft-recipient leads to a Microsoft 365 cohort only, kill the cross-routed batch.
- If all the above is clean → strip the Outlook cohort from the active campaign, age the domains another 30d, swap to fresh Microsoft 365 inboxes (hot-swap recovery, SOP-1).
F.4 Caveat
The 0+0+200 pattern can false-positive on a campaign with a bad list hitting the same shape. The placement test in F.2 disambiguates.
Runbook G - "550 5.7.708 dirty IP pool"
G.1 Identify
Bounce text contains 550 5.7.708 from Microsoft sources. This is Microsoft's "dirty IP pool" code - your sending IP is temporarily blocked due to reputation issues at the IP level, NOT your individual domain.
G.2 Critical distinction
This is NOT a per-domain blacklist (Runbook C) and NOT a per-mailbox warmup block (Runbook D). It's a shared-IP-pool reputation event - it affects every domain sending through that IP regardless of their individual reputation. On a shared-IP provider (most cheap inbox tiers), one bad neighbor can drag you into 550 5.7.708 territory through no fault of your own.
G.3 Fix
- Pause the affected cohort immediately. Continuing to send into a Microsoft block compounds the reputation problem.
- Contact your infra provider. Open the Microsoft support ticket through them - the IP-pool issue is theirs to resolve with Microsoft, not yours.
- Reduce volume 75% on resume.
- Long-term: consider migrating to a dedicated-IP tier (dedicated > shared at any premium worth paying once this happens).
G.4 Log the incident
Track frequency per provider. If you see >2 incidents/quarter from the same provider, that provider's shared-IP pool is contaminated and warrants leaving.
SOP-1 - Hot-swap recovery
When an Outlook/Proofpoint cohort underperforms and you need to keep the campaign sending while you fix:
- Strip the campaign to Gmail-only sending temporarily (remove all Microsoft 365 inboxes from the campaign's account assignment). Maintains send volume during recovery.
- Pull aged Outlook domains from a 4-8 week aging pool (the pre-aged domain reserve).
- Provision new Microsoft 365 inboxes on aged domains + a 30d Microsoft warmup.
- Re-launch the Outlook cohort on the new (now-warmed) inboxes after the 30d warmup completes.
- Old Outlook domains → cooling pool (if <90d, candidate for repair) or kill. Never reuse burned-cohort copy on the new inboxes.
Prereq: the aging pool must exist. If you haven't been provisioning aged domains in reserve, the hot-swap can't fire - fall back to the full Runbook C replace timeline.
SOP-2 - Preemptive domain kill
Triggered when ≥20% of the inboxes on a single domain drop below 98% reputation.
- Confirm via a lightweight placement test from 1 mailbox on the suspect domain. Domain in spam → proceed. Inbox-clean → skip the kill, investigate inbox-level issues.
- Flag and confirm with a human - never auto-execute (domain-wide blast radius too high).
- Cancel subscriptions FIRST, then delete from the sequencer. Reverse order = zombie billing.
- Replace from the warming pool.
- Audit the upstream cause - why did 20%+ of inboxes degrade in tandem? Usually a shared copy fingerprint, a shared bad list segment, or a shared IP-pool issue (Runbook G).
SOP-3 - Microsoft 30-day warmup
For any NEW Microsoft inbox:
- Days 1-30: pure warmup only, zero cold.
- Day 30 launch: 2-3 cold/day per inbox + warmup continues at 1:1.
- Day 38+: full ~5 cold/day per inbox at ~25 inboxes/domain steady state.
Microsoft is far less forgiving than Google on young inboxes sending cold - this ramp is the difference between a usable cohort and a burned one.
72-hour total-failure checklist
When multiple runbooks trigger at once or an account is at ~0 reply rate:
- Pause all campaigns for the affected account.
- Audit suite - re-run Phase 1 of this skill end-to-end.
- Spam test on 2 sender subsets - lightweight placement test, active cohort vs warming.
- DNS auth verify all sending domains (
references/diagnostics.md). - Provider IP-pool health - contact the provider, check their status page, blacklist-check the assigned IPs.
- Reduce volume 75% - restart at 25% of pre-incident.
- Use known-good copy - the last version with a confirmed >2% reply rate. Do NOT launch new copy during recovery.
- Daily watch for 7 days - reply rate, bounce, placement. Alert if any threshold trips again.
When to escalate / give up
- In spam >2 weeks despite fixes → infra-level. Consider migrating sending platforms.
- Multiple domains permanently blacklisted → IP-pool issue. Switch providers.
- Domain reputation never recovers after 4 weeks → retire, replace.
Cold email infrastructure is disposable. A domain costs a few dollars. Your time is worth more than preserving a burned sender.
After the incident
- Wait 7 d post-restart. Re-run Phase 1 to confirm recovery.
- Reply rate back to baseline AND ≥1% on ≥200 sends.
- If domains/inboxes were replaced, expect 2 weeks of warmup before full recovery.
- Document: cause, fix, time-to-recovery. If a new symptom or fix pattern emerges, add it to this skill.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

