agentleFS
Insightful-PipeInsightful-Pipe/email-marketing-skills/skills/competitor-email-teardown/SKILL.md

competitor-email-teardown

Skill1 starsChanged 16 days agoMITView on GitHub

Benchmarks rival email programs from inbox exports, measuring welcome speed, frequency, signup offers, discounts, subject lines and sale timing against your Klaviyo or Mailchimp program, with three sized tests. Use when asked to analyze, track or tear down competitor emails or welcome series.

The file

Show its rules and text

We split this skill into the rules it gives an agent

16 rules. About 0 are common, 12 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈12

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Ask whether the user already has rival emails with dates.

    line 14
  2. Distinctive:

    Leave the inboxes alone. Don't buy anything and don't enter payment details.

    line 19
  3. Distinctive:

    Wait 56 days if you can, ideally across one sale event (a holiday, a seasonal sale).

    line 20
  4. Distinctive:

    When the teardown is done, unsubscribe the research addresses.

    line 21
  5. Distinctive:

    The window and time zone: Default: day 0 is the signup date, through day 55, in the user's time zone.

    line 28
  6. Distinctive:

    Ask for every email each inbox received in the window, as a paste, a CSV, or a mailbox export (Gmail: put each rival under its own label, …

    line 35
  7. Very specific:

    Call query_contexts with request="accounts" for klaviyo or mailchimp.

    line 46
  8. Very specific:

    Call query_contexts with request="actions_details" for the actions below and follow the body shapes it returns.

    line 47
  9. Very specific:

    Every read goes through query_data.

    line 49
  10. Distinctive:

    If a call fails, mark that part unknown, name the call, and never fill the gap with a guess.

    line 53
  11. Distinctive:

    No seed inbox export?: Read what your own seed profile received instead:

    line 76
  12. Very specific:

    Don't pass fields[event] with profile_id.

    line 80
  13. Distinctive:

    Frequency is not a test here: When your regular emails a week are 0.5 or more away from the rival median, list it as a handoff: a frequency …

    line 148
  14. Distinctive:

    Never sign up, subscribe or create an inbox for the user, and never use a real person's details for research.

    line 171
  15. Distinctive:

    Every number in the report comes from the inboxes or from a call this skill made.

    line 174
  16. Distinctive:

    No em or en dashes in the report.

    line 175

≈ “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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Competitor Email Teardown

Turns a few weeks of rival emails into numbers you can set beside your own program. Every measure has one fixed definition, so a rival's inbox and yours are counted the same way. Your own side adds what no inbox can show: your click and order rates and your signup volume, which decide which gaps are worth a test and how long each test takes.

Before you start

The research inboxes (the user does this, never Claude)

Ask whether the user already has rival emails with dates. If not, give them this setup and stop until they have at least 28 days of mail:

  1. Make one address per rival on a domain or mailbox the business owns, for example research+rivalname@yourbrand.com. Never use a person's private address, a colleague's name, or anyone else's details.
  2. Sign up to every rival on the same day, from their own signup form, and write down the exact time of each signup and the offer the form showed.
  3. Sign your own seed address up to your own list the same day, the same way. Your program is then measured exactly like theirs.
  4. Leave the inboxes alone. Don't buy anything and don't enter payment details. Starting a cart with the research address is optional; if the user does it, note which rival and when, because that triggers cart emails the comparison would otherwise miss.
  5. Wait 56 days if you can, ideally across one sale event (a holiday, a seasonal sale). 28 days is the minimum.
  6. When the teardown is done, unsubscribe the research addresses.

Claude never fills in a signup form, never creates an inbox and never subscribes anyone.

What to ask

  1. The rivals. Two to five, with their sites.
  2. The window and time zone. Default: day 0 is the signup date, through day 55, in the user's time zone.
  3. The sale event inside the window, with its date. Default: none; the sale rows are then "not graded".
  4. The discount ceiling: the deepest discount the business will ever give. Default: the deepest % off in the user's own emails in the window.
  5. The smallest lift worth detecting in a test. Default: 20% relative.

The emails, without a connection

Ask for every email each inbox received in the window, as a paste, a CSV, or a mailbox export (Gmail: put each rival under its own label, then export with Google Takeout as MBOX; Outlook: export the folder). Each email needs:

  • received date and time, with the time zone
  • sender name
  • subject and preview text
  • the body text, or the first lines of it

Plus, per program: the signup time and the offer the signup form showed. For the user's own results without a connection, ask for the welcome or onboarding report for the last 90 days (recipients, click rate, orders or conversions per email) and the campaign report for the same 90 days.

How to run the queries

With the InsightfulPipe MCP connected:

  1. Call query_contexts with request="accounts" for klaviyo or mailchimp. Note workspace_id, brand_id and the account. If there are several, ask which one is the user's own.
  2. Call query_contexts with request="actions_details" for the actions below and follow the body shapes it returns.

Every read goes through query_data. Rules found on a live account:

  • An empty answer is not an error. An account with no flows or campaigns answers success with an empty list. Say "no history" for that part.
  • Klaviyo rates are fractions (0.0125 is 1.25%).
  • Never fire calls in parallel. On a 429, wait the retry_after seconds and retry once.
  • If a call fails, mark that part unknown, name the call, and never fill the gap with a guess.

Your own program: Klaviyo

Which automations exist:

{"action": "get_flows", "fields[flow]": "name,status,trigger_type", "auto_paginate": true}

Welcome and other flow results per email, last 90 days:

{"action": "get_flow_report", "body": {"data": {"type": "flow-values-report", "attributes": {"timeframe": {"key": "last_90_days"}, "filter": "equals(send_channel,\"email\")", "group_by": ["flow_id", "flow_message_id", "flow_name", "flow_message_name"], "statistics": ["recipients", "click_rate", "unsubscribe_rate", "conversions", "conversion_rate"]}}}}

Campaigns sent, and their results for the same 90 days:

{"action": "get_campaigns", "filter": "and(equals(messages.channel,'email'),equals(status,'Sent'))", "fields[campaign]": "name,send_time", "sort": "-created_at", "auto_paginate": true}
{"action": "get_campaign_report", "body": {"data": {"type": "campaign-values-report", "attributes": {"timeframe": {"key": "last_90_days"}, "group_by": ["campaign_id", "campaign_message_id", "send_channel"], "statistics": ["recipients", "click_rate", "unsubscribe_rate", "conversion_rate"]}}}}
  • get_campaigns fails without the channel filter, and status values are Title Case.
  • The report actions count conversions on the account's Placed Order metric unless you pass conversion_metric_id; meta.report.conversion_metric names the one used.
  • Flow delays are not in these actions. Your welcome timing comes from the seed inbox.

No seed inbox export? Read what your own seed profile received instead:

  1. get_metrics (auto_paginate: true), and keep the id of Received Email.
  2. get_profiles with "filter": "equals(email,'<SEED_ADDRESS>')". Use only the user's own test address, never a customer's.
  3. get_events with "filter": "and(equals(metric_id,\"<RECEIVED_EMAIL_ID>\"),equals(profile_id,\"<PROFILE_ID>\"))", "sort": "datetime", "auto_paginate": true.
  • Don't pass fields[event] with profile_id; the call fails (allowed fields are datetime, event_properties, id, timestamp, uuid). An unknown profile id returns "does not exist".
  • Each event gives the time, and its event_properties carry the subject (Klaviyo names it Subject; check the first event's keys). Preview text and body are not there, so offers inside the body need the inbox export. Say so in "Not captured".

Your own program: Mailchimp (untested until a Mailchimp account is connected)

  • list_automations, then list_automation_emails with each workflow_id: each email's delay and subject.
  • list_campaigns with status sent and since_send_time 90 days back: send_time, settings.subject_line, settings.preview_text, recipients.recipient_count.
  • list_reports with the same since_send_time: emails_sent, clicks.click_rate, unsubscribed.
  • Customer Journeys built in Mailchimp's newer editor may not appear in list_automations. Then the seed inbox is the only timing source.

The rivals' public signals

Signup offer on the page. The site crawler's text checkers return nothing on many storefronts (bot walls, popups drawn by scripts), so take a picture instead:

{"platform": "screenshot", "action": "capture", "url": "https://www.rival.example", "include_desktop": false, "include_mobile": true, "modes": ["mobile"]}

It returns data.captures.mobile.public_url, a full-page image. Read the signup offer from it. The capture can render in another country's currency and may not show a popup; if no offer is visible, use what the user saw when they signed up.

Signup offer in paid ads (optional). ads-libraries:

  • facebook_search_companies with query (the rival's name) and country US, to get the page_id. Pick the verified page that matches the site.
  • facebook_company_ads with page_id, country, max_ads 30 and trim true. Rows are under data.results; start_date is in Unix seconds. Catalog ads have empty body_text, and the call can return more rows than max_ads.
  • Count the active ads whose text offers something on the first order. A rival paying to advertise its signup offer is telling you the offer earns its keep.

The measures

Count every program the same way. Day numbers are calendar days after the signup date in the window's time zone; the signup day is day 0.

MeasureDefinition
Minutes to first emailfirst received time minus the signup time
Emails in the first 14 daysevery email received on days 0 to 13
Regular emails a weekemails on days 14 to the window's last day that don't name the sale event, divided by the number of weeks in that span
Offerthe email offers N% off, $N off, free shipping, a free sample, a free gift, 2 for 1 or buy one get one, anywhere in the subject, preview or body
Typepromotional if it has an offer; otherwise product if it announces something new, back in stock, a new harvest or now available; otherwise content
Signup offerthe offer in the first email: a % off, a non-discount offer (such as a free sample), or none
Deepest % offthe highest % off in any email in the window
Climbing welcome offeramong days 0 to 13, the emails that carry a % off and mention a code or the first order: yes when a later one offers more than an earlier one
Subject patternsmedian length in characters, and the share of subjects with a number, an offer, an urgency word (end, ends, ending, last chance, tonight, final hours, today only, hours left), a question mark, an emoji
Sale timingfor emails naming the sale event: how many days before the event the first one arrived, how many there were, the most in one day, the deepest offer, and whether one said "extended" after an earlier sale subject used an urgency word
Usual sendthe most common weekday and the most common send time, sale emails left out

The rival median is the median across rivals; the user's own program is never part of it.

Decisions

Adopt

Adopt only a change that costs no margin, adds no regular sends, and that at least 2 rivals share:

  • Welcome speed. Your first email takes more than 60 minutes and at least 2 rivals send theirs within 10: send email 1 on signup.
  • Sale lead time. Your first sale email arrives less than 2 days before the event and at least 2 rivals start 2 or more days ahead: announce the next sale to subscribers at least 2 days before.

Avoid

List a rival pattern under "avoid" when the inbox shows:

  • A deceptive subject: starts with Re: or Fwd: with no earlier thread, or claims an order or account action that never happened. US CAN-SPAM forbids deceptive subject lines.
  • A deadline that moves: a sale said it was ending, then came back as extended.
  • A discount above the user's ceiling, with the climbing welcome offer named when there is one.

Test candidates

GapBecomes a candidate whenMetricUnit and volume
Subject styleyour median subject length is 10 or more characters away from the rival mediancampaign click rateone campaign recipient; half the average campaign recipients per arm per send
Welcome lengthyou send at least 2 fewer emails in the first 14 days than the rival medianwelcome orders per new subscriber (welcome flow conversions ÷ welcome email 1 recipients)one new subscriber; half the weekly new subscribers per arm
Signup offer depthyour signup % off is 5 or more points away from the rival median (a rival with no discount counts as 0)welcome email 1 placed-order rate, then margin per new subscriberone new subscriber; half the weekly new subscribers per arm

New subscribers a week = welcome email 1 recipients over the last 90 days ÷ (90 ÷ 7), rounded. Use click rate, not open rate: Apple Mail Privacy Protection opens almost every email it receives.

Frequency is not a test here. When your regular emails a week are 0.5 or more away from the rival median, list it as a handoff: a frequency change plays out per person over weeks, and it needs your own unsubscribe and complaint history.

Size each test: n per arm = ⌈(1.959964 + 0.841621)² × (p1(1 − p1) + p2(1 − p2)) ÷ (p2 − p1)²⌉, with p1 the baseline from your own data and p2 = p1 × (1 + the smallest lift). That is 5% significance, two-sided, at 80% power. Time to read = n ÷ your volume per arm, rounded up to whole sends or weeks. A test that needs more than 12 weeks is marked slow; say what lift it could read in 12 weeks instead.

Pick three. Rank candidates by how many rivals sit on the other side of your value, then by the units affected each week. The top three are the tests; the rest are "later".

The report

  1. Summary: the window, the emails read per program, and the two biggest gaps.
  2. Side by side: one row per measure, a column per program, and the rival median.
  3. Rival profiles: per rival, its signup offer, what the signup page said, how many active ads push the signup offer, its usual send, its sale timing, and a two-line read of its program.
  4. Your results: live and draft flows, new subscribers a week, welcome and campaign click and order rates.
  5. Adopt, avoid, three tests and handoffs, each with the numbers behind it.
  6. Not captured: what the inboxes couldn't see, such as cart flows that were never triggered and every rival's own results.

Rival patterns show what rivals do, not that it works for them. Say so once, and let the tests decide.

What this skill changes

Nothing. It reads, measures and recommends. It never edits a flow, a campaign or a signup form, and never sends or schedules an email.

Rules

  • Never sign up, subscribe or create an inbox for the user, and never use a real person's details for research.
  • Quote rival subject lines as evidence; never republish a rival's email body or images.
  • Count every program with the same definitions and the same window.
  • Every number in the report comes from the inboxes or from a call this skill made.
  • No em or en dashes in the report.

Go deeper

  • klaviyo-flow-auditor, if installed: audit your own flows end to end.
  • email-sequence-builder, if installed: write the extra welcome emails a test calls for.
  • email-campaign-calendar, if installed: set a send frequency from your own fatigue history.
  • competitor-swot-analyzer and competitor-ad-research, if installed: the rest of each rival's marketing.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.