Email Subject Line Tester

Email subject line optimization agent. Generates 10 subject line variants for any email, scores each by open rate predictors (urgency, personalization, curio...

Security Scan

The skill's requirements and instructions are internally consistent with an email subject line testing tool; it is instruction-only, requests no credentials, and its file writes to ~/email-subjects/ fit its stated purpose.

Purpose & Capability

Name, description, and commands align with an email subject line optimization tool. No unrelated credentials, binaries, or install steps are requested.

Instruction Scope

SKILL.md confines operations to generating/scoring subject lines and saving session data. It instructs the agent to create ~/email-subjects/ and write memory.md, history/, and benchmarks.md — which is consistent with the stated workspace behavior but means user-provided or pasted content (including any PII) will be stored locally.

Install Mechanism

No install spec or code files are present (instruction-only), so nothing will be downloaded or written during installation beyond the agent executing its instructions at runtime.

Credentials

No environment variables, credentials, or external integrations are required. The absence of secrets is proportionate to the declared functionality.

Persistence & Privilege

always is false and the skill does not request elevated platform privileges. It does request the agent use Bash (allowed-tools), which allows reading/writing files when the skill runs — appropriate for saving history but worth noting since it enables filesystem writes to the user's home directory.

Assessment

This skill appears coherent for subject-line generation and A/B test guidance. Before using it: (1) review and sanitize any pasted competitor or audience data you don’t want stored, since the agent will save sessions and profiles to ~/email-subjects/; (2) be aware that allowed-tools: Bash permits filesystem operations — verify what files are created and their permissions; (3) confirm the spam-check and benchmark logic are internal (no external calls) if you want to avoid sending data to third parties; (4) optionally inspect the referenced GitHub repo before enabling, and run the agent in an environment where writing to your home directory is acceptable.

AI-powered email subject line optimization agent — generates 10 variants, scores each on 7 open-rate predictor signals, and selects the top 3 for A/B testing.

Describe your email's topic, audience, and goal. The agent produces scored variants across multiple psychological angles (urgency, curiosity, personalization, social proof) and gives you a ready-to-launch A/B test plan with sample size guidance.

Commands

subject test <topic>               # generate and score 10 subject line variants for a topic
subject generate                   # generate variants with more context (audience, goal, tone)
subject score <line>               # score a specific subject line you already have
subject ab test                    # structure a full A/B test plan with sample size formula
subject analyze competitors        # analyze subject lines from competitor emails you paste
subject by industry                # get industry-specific benchmarks and top-performing patterns
subject history                    # show previously tested subject lines and their scores
subject save                       # save current session results to ~/email-subjects/

What Data to Provide

The agent works with:

• Topic description — "promotional email for 30% off summer sale, audience is past buyers"
• Draft subject lines — paste your own for scoring and improvement suggestions
• Competitor examples — paste subject lines from competitor emails for pattern analysis
• Audience details — industry, demographic, relationship (subscriber, buyer, lead), engagement tier
• Email goal — promotional, transactional, re-engagement, newsletter, event invite

No integrations required. Works entirely from your descriptions.

Workspace

Creates ~/email-subjects/ containing:

• memory.md — saved audience profiles, brand voice notes, and past A/B test results
• history/ — past testing sessions saved as markdown (session-YYYY-MM-DD.md)
• benchmarks.md — industry benchmark reference updated during sessions

Analysis Framework

1. The 7 Open-Rate Predictor Signals

Each subject line is scored 0-10 on each signal; total score is out of 70:

Signal 1 — Urgency Words

• Time-limited language: "today only", "ends tonight", "last chance", "24 hours left"
• Quantity scarcity: "only 3 left", "limited spots", "while supplies last"
• Diminishes with overuse — flag if brand history shows urgency fatigue

Signal 2 — Personalization Tokens

• Name token {{first_name}} adds 2-5% open rate lift on average
• Behavioral personalization: "You left something behind", "Based on your last order"
• Segment-specific language (buyer vs. subscriber vs. VIP)

Signal 3 — Question Format

• Open questions create curiosity loops: "What's your biggest email mistake?"
• Yes/No questions drive agreement priming: "Ready to double your open rates?"
• Rhetorical questions require no answer — lower friction than calls to action

Signal 4 — Number Inclusion

• Specific numbers outperform vague claims: "Save $47" beats "Save money"
• Odd numbers slightly outperform round numbers in most studies
• List-format subject lines: "5 mistakes killing your open rates"

Signal 5 — Emoji Presence

• Single relevant emoji adds novelty in crowded inboxes; more than 2 reduces credibility
• Emoji at start of subject performs differently than at end (test both)
• Inappropriate for B2B enterprise, legal, financial contexts — flag by industry

Signal 6 — Character Length

• Optimal range: 30-50 characters for desktop and mobile rendering
• Under 20 characters: too vague, loses context
• Over 60 characters: truncated on mobile (58% of opens are mobile)
• Preheader pairing: subject + preheader combined should tell the full story in 90 characters

Signal 7 — Power Words

• High-engagement triggers: "exclusive", "secret", "proven", "free", "new", "you"
• Spam-trigger words to avoid: "100% free", "act now", "cash bonus", "no cost", "winner"
• Run spam filter check on every generated variant

2. Industry Benchmark Reference

Industry	Average Open Rate	Top Quartile
Ecommerce	15-20%	>25%
SaaS / Software	20-25%	>32%
Newsletter / Media	25-35%	>45%
B2B Services	20-28%	>35%
Nonprofit	26-30%	>40%
Healthcare	22-27%	>35%

3. Spam Trigger Detection

• Scan each variant against known spam trigger word list
• Flag phrases that increase spam folder placement risk
• Check for ALL CAPS usage (more than 2 consecutive caps words triggers filters)
• Check for excessive punctuation (!!! or ???)

4. Mobile Preview Check

• Simulate rendering at 40 characters (iPhone lock screen) and 58 characters (Gmail mobile)
• Flag subject lines that truncate at an awkward word break
• Suggest preheader text that completes the message naturally when subject is truncated

5. A/B Test Setup Guidance

• Minimum sample size formula: n = (Z^2 × p × (1-p)) / E^2
  • Z = 1.96 for 95% confidence, p = baseline open rate, E = minimum detectable effect (typically 0.02)
  • Example: 25% baseline, detect 2pp lift → n = 2,401 per variant
• Test only one variable per test (subject line only, never combine with send time changes)
• Recommended test split: 20% / 20% test, 60% winner send
• Minimum test duration: 4 hours before declaring winner (allow for time-zone spread)

Output Format

Every subject test run outputs:

1. 10 Scored Variants — each with total score /70, per-signal breakdown, and character count
2. Top 3 Picks — recommended for A/B testing, with rationale for each selection
3. Spam Flag Report — any variants with trigger words highlighted
4. Mobile Preview Simulation — truncated rendering at 40 and 58 characters
5. A/B Test Plan — test setup instructions with sample size recommendation
6. Preheader Suggestions — paired preheader for each top-3 variant

Rules

1. Always generate exactly 10 variants before scoring — never fewer
2. Never recommend a variant containing known spam trigger words without flagging the risk
3. Score every variant on all 7 signals — no signal may be skipped
4. Flag when the audience or industry context makes certain signals inappropriate (e.g., emoji in B2B financial services)
5. Always include character count and mobile truncation preview for every variant
6. When scoring a user-provided subject line, explain each signal score individually — not just the total
7. Save session results to ~/email-subjects/history/ when the user requests subject save