Email review

Email Review Skill

Review email copy against critical copy rules, QA checklists, and client preferences. Output a scored review and, if quality is sufficient, produce emails_v2.md as the final reviewed version.

Execution Tasks

Execute these tasks in order. Do NOT skip any task unless noted.

Task 1: Load Context

• Read {segment_folder}/emails.md - The email copy to review
• Read {segment_folder}/strategy.md - Campaign strategy and goals
• Read {workspace}/research.md - ICP, value proposition, messaging themes, product info, case studies ({workspace} = the client campaign workspace folder (the orchestrator passes this path))
• Read {workspace}/notes.md (if exists) - Client-specific preferences (HIGHEST PRIORITY)

Task 2: Load Review Reference Material

• Read qa.md from this skill folder - QA checklists and scoring criteria
• Read extra_info.md from this skill folder - Exercises, tips, common mistakes
• Read ../email-copywriting/references/gold-standard.md - the 9/10 bar the copy is scored against
• Read ../email-copywriting/references/spam-words.md - tiered spam-word policy (NEVER/HIGH/MEDIUM + budget rule)
• Read ../email-copywriting/references/sound-human.md - AI-tell lexicon/structures and the Voice Card spec
• Find the ICP Voice Card in the ## Notes section of emails.md - FLAG its absence as a violation

Task 3: Review Against Critical Copy Rules Checklist

MANDATORY: Check EVERY email against these non-negotiable rules.

Transitions & Flow (CRITICAL - top failure mode)

• No abrupt transitions from {{First Line}} to body - personalized opening flows naturally into next sentence
• FLAG any email where the static body's first sentence ignores the personalized line and jumps into a generic statement (e.g. straight into "Scaling a SaaS company..."). The body's opening must read as the next beat of the {{First Line}}, not a topic change.
• Transitional context provided or first line topic connects to body

Language to Avoid (REJECT if found)

• No "Quick followup", "quick question", "touching base"
• No "No worries if timing isn't right" or apologetic/defensive language
• No "Quick insight:" prefix - state insights directly
• No typical sales clichés that scream "cold email"
• Final-email salesy endings (REJECT) - the last email must NOT announce itself as a follow-up or last attempt. REJECT "Last note from me", "Last email", "Just following up", "Following up", "Circling back", "Checking in", "Wanted to follow up", "Bumping this", "One last try". This is the salesy ending operators flag most - check the final email specifically, every time.

Spam Words - Budgeted (count and report per email - HARD)

Policy and full tiers in ../email-copywriting/references/spam-words.md.

• Count every spam word per email and report word (tier) in the review output.
• NEVER tier = REJECT: "free", "credit card", phishing vocabulary (verify your account, action required...), fake Re:/Fwd:, scam cluster (winner, no obligation, risk-free, 100% guaranteed...).
• HIGH tier: more than one HIGH word in an email = REJECT; even one = FLAG with the safe alternative from the reference table.
• MEDIUM tier: more than 2 in one email = REJECT; two from the same cluster in one sentence = FLAG.
• Subject lines: any tiered word, $, %, or stacked numbers = REJECT.
• Structural triggers: ALL CAPS, more than one !, 2+ links, URL shorteners, attachments = REJECT.
• No false/unverifiable offer claims - e.g. "no card needed" when a card is actually required, or implying a trial is unlimited when it isn't. Cross-check every offer claim against {workspace}/research.md; FLAG anything you cannot verify.

Value-Prop Compression (the 9/10 bar - HARD)

Scored against gold-standard.md:

• The one-breath test: Email 1 must contain a single plain sentence carrying the whole offer (deliverable + who + what they get). If the offer is smeared across three sentences, REJECT with the compressed rewrite.
• Repeatability: after one skim, could the prospect repeat the deal to a colleague in their own words? If they'd say "some [category] thing", the value prop failed - show a rewrite.
• Dead-sentence scan: delete each sentence mentally; FLAG every sentence whose removal changes nothing (industry observations, restated pain, throat-clearing).
• Concrete > abstract: FLAG hype adjectives and abstract benefit language; every number must be real and specific.
• Shown, not claimed: risk reversal as specific consequence beats "guarantee"; scarcity as a plain true reason beats "limited spots".

One-on-One Favor Framing (REJECT marketing emails - HARD)

The rendered email must read like one human typing a note to another to do them a favor - not a marketing email. Frame + tests in ../email-copywriting/references/sound-human.md.

• The list test: if the email would read the same sent to 500 people unchanged, REJECT - the static body must read like it was typed for this one reader.
• The favor test: mentally delete every sentence about the sender's product/company; if nothing is left for the reader, REJECT. The email must hand the reader something - a free deliverable, a fix for a problem they have, information they'd want anyway.
• No feature/brochure framing: REJECT feature lists and "we offer / our platform provides / our solution" framing. What THEY get in plain verbs; mechanism gets one clause at most.
• We/you balance: FLAG any email where sender-focused sentences outnumber reader-focused ones.

Sound Human / AI-Tells (REJECT if found)

Full lexicon and structures in ../email-copywriting/references/sound-human.md.

• No AI-tell vocabulary: delve, leverage, streamline, seamless, robust, unlock, elevate, empower, transform, journey, landscape, game-changer, tailored solution, actionable insights...
• No dead cold-email phrases: "I hope this email finds you well", "I came across", "I couldn't help but notice", "I was impressed by", "Congratulations on your recent funding" (as formula), "As a [title], you know..."
• No AI-tell structures: "It's not just X, it's Y" / "not only... but also"; tidy triads ("fast, reliable, and affordable"); three consecutive same-length sentences; "serves as"/"stands as"/"boasts" copula dodging; hedge chains; compliment-then-pitch seams.
• Passes the read-aloud test: FLAG any phrase a real person would never say to the prospect's face.
• Human signals present: contractions, varied sentence rhythm, plain first-person verbs; at most ONE personality flourish per email.
• Do NOT flag deliberate small imperfections - a lowercase start, a missing comma, casual shorthand ("btw", "thats") at up to ~1 per email / ~2 per sequence is intended human roughness (spec in sound-human.md). Do NOT "correct" them in emails_v2.md. DO reject any imperfection in the prospect's name, their company, the client's company, links, numbers, or offer claims; FLAG scrambled-letter fake typos ("teh", "recieve") and more than ~2 imperfections per sequence.

ICP Voice Match

• Voice Card exists in emails.md ## Notes (REJECT if missing) and is derived from client context, not generic.
• Copy matches the card: vocabulary the ICP uses daily, none of the card's "never say" phrases, formality level consistent with how the ICP talks to peers.
• Per-segment re-voicing: if this segment's copy differs from a sibling segment only by an industry-noun swap, FLAG it - the offer must be re-framed in this ICP's terms, not re-labeled.
• Proof type matches the ICP: the evidence offered is the kind this reader believes (peer names, benchmarks, revenue math - per the card).

Greetings (Check EVERY email, including follow-ups)

• Every email opens with a greeting followed by an empty line. Email 1: Hey {firstName},. Follow-ups: Hey {firstName}, OR the bare {firstName}, (both valid - the bare form reads like a continuing conversation; do NOT flag it)
• FLAG follow-ups that jump straight into the body with no greeting - threading off Email 1 does not excuse a missing greeting

Variables & Signature (Format - REJECT)

• AI variables use double braces {{Title Case}} - REJECT any [square bracket] AI variable ([First Line], [PS Line], [Subject Line]). The double-brace form is what renders correctly in the app on push.
• One AI variable = one value per LEAD, not per email - REJECT unintended reuse. The same body token in multiple emails renders the IDENTICAL text in each (e.g. {{First Line}} in Emails 1-3 = the same sentence sent three times). Each email's opener must be a distinct-named variable ({{First Line}}, {{Bump First Line}}, {{Nudge First Line}}) with a matching H3 in prompts.md. {{Subject Line}} reused in every subject is intentional and correct - do not flag it. Diagnostic: in the approval portal, a highlighted variable filled on Email 1 but blank on later emails means REUSED, not missing.
• Every email signs off with the sender-name signature block - REJECT {sender_signature} or any signature missing {sender_first_name} {sender_last_name} - [title] at [linked Client Company].
• Lead variables use {camelCase} - {firstName}, not {first_name}.
• PS variable ({{PS Line}} / {{Final PS Line}}) sits on its own line after the signature block.

Formatting Requirements

• Short paragraphs only: 1-3 sentences max per paragraph
• Adequate spacing between thoughts (remember: AI adds 1-2 sentences)
• Never ALL CAPS for emphasis (use structure or italics only)

Brevity & Punch (HARD - cold emails are almost always too long; the #1 operator complaint)

• Count the static body words of every email (body = prose + CTA, excluding the greeting, the {{...}} AI-variable lines, and the signature block; a static PS is also excluded but capped at ~35 words in at most 1-2 emails). Hard ceiling 80 words. REJECT any email over 80 - this is a fail, not a flag - EXCEPT one value-stack email per sequence, which may reach ~100 words if the overage is a numbered list where every line carries a real number (gold-standard.md Email 2). Follow-ups should land well under (target ≤55; closes 35-45). Report the per-email count.
• Run the "cut at least 30%" pass on every email - if it can be cut ~30% without losing meaning, it is bloated: REJECT and show the tighter version.
• Prefer short sentences - flag long, multi-clause sentences.
• Punchy, not staccato - flag jittery one-sentence-per-line fragments ("It just runs."); copy should keep full thoughts grouped into 1-2 short paragraphs.

Sequence Length

• Default is 3 emails, not 4-5 - if the sequence has 4+ emails, verify each extra email carries a distinct worth-sending idea. FLAG any follow-up that just restates an earlier email and recommend cutting it.

First Email Requirements (Email 1 ONLY)

• MUST mention the solution - not just the problem
• States problem AND offers solution in the same email
• Don't ask if they have problems - state their problem and offer solution
• UVP stated clearly - don't keep value prop secret

CTA Requirements (Check ALL emails - HARD)

• Every email's CTA asks for the real offer, including Email 1. REJECT any email whose close does not ask for the actual offer (the call, demo, walkthrough, or sample). The first email must carry the strongest offer CTA, not a soft warm-up.
• REJECT diagnostic / soft questions used as the CTA - "Do you have a system for X?", "Do you have anything in place?", "Where do your clients come from?", "Is it worth a look?", "a quick look", "Curious if...?", "Worth exploring...?". Asking about their situation is not a CTA.
• Strong and direct CTAs - a direct ask for the offer ("Want me to walk you through it?", "Open to a quick walkthrough?"), or positioned as a "strategy call" / "conversion optimization call" for SaaS
• No "Should I close the loop?" - use direct unsubscribe language instead
• No complex rhetorical questions that require too much thought to answer
• No time commitments in CTA - duration details ("30 minutes", "15 min") should not appear in the CTA line
• No keyword trigger CTAs - "Reply 'X' and I'll send Y" patterns are spam signals; use plain "reply" CTAs instead
• Question + link-on-its-own-line is PREFERRED, not a violation - a short direct question ("Do you want to apply?", "Want a seat?") with the link on the next line by itself is a strong CTA pattern. Do not flag it for "link not in a sentence"; flag only if the question is weak or the link is embedded/buried instead.

Rhetorical Questions

• If rhetorical question used, it's immediately followed by answer/solution in next sentence
• Never leave prospects hanging - rhetorical questions need payoff

Offer Strategy (Mid-Market)

• Sells the PRODUCT, not audits/analysis/friction analysis
• Solution-focused for mid-market - not indie hacker-style audits
• Analysis/audit mentioned as proof/demo only, not primary offer

Case Study Relevance

• Proof points match prospect's vertical - no irrelevant case studies
• No B2C examples for B2B prospects (e.g., no Crocs for B2B SaaS)
• Same or adjacent industry case studies used
• If no relevant case study, use generic metric without company name

Subject Line Threading

• Every email uses an AI-generated subject line {{Subject Line}} - operator standard (Cristian, 2026-06-03): all steps carry {{Subject Line}}, including follow-ups. Do NOT leave follow-up subjects empty and do NOT flag a follow-up that has {{Subject Line}}. (The parser emits {{Subject Line}} for every step via _default_subject_line.)

Task 4: Run QA Checklist Scoring

• Score each email using criteria from qa.md
• Run self-review questions (assume you're a cold prospect)
• Check strategy alignment

Task 5: Run Quality Exercises

• "So What?" Test - For every line, ask "So what?" If you can't justify it, cut it
• Word Reduction - Identify lines that can be tightened (strip unnecessary words)
• Read Aloud Test - Check for awkward phrasing
• Feature-to-Story Transformation - Flag any boring feature dumps

Task 6: Output Review

Generate a structured review with:

## Email Review

### Overall Score: [1-5, can be decimal like 4.5]

### Critical Copy Rules Violations
[List any violations found in Task 3, or "None found"]

### Email-by-Email Feedback

**Email 1: [Title]**
- Score: [1-5]
- Issues: [list specific problems]
- Suggestions: [specific improvements with before/after]

**Email 2: [Title]**
...

[Repeat for each email]

### Client-Specific Notes Applied
[Any notes from {workspace}/notes.md that were applied]

### Quality Exercise Results
- "So What?" failures: [lines that couldn't be justified]
- Tightening opportunities: [lines that can lose words]
- Awkward phrasing: [flagged from read aloud test]
- Length & brevity: [per-email word counts; any email that fails the "30% shorter?" pass, with the tighter rewrite]
- Spam-word budget: [per email: each spam word found as `word (tier)`, or "clean"]
- One-breath test: [the offer sentence found in Email 1, or REJECT + compressed rewrite]
- One-on-one favor check: [per email: does it pass the list test and favor test? any brochure/feature framing or we-heavy sentences]
- AI-tell scan: [any lexicon words, dead phrases, or structures found, per email]
- Voice Card match: [card present? copy in the ICP's language? any "never say" phrases used?]
- Greeting check: [any follow-up missing a greeting (`Hey {firstName},` or bare `{firstName},`)]
- First-line bridge: [any email where the body's first sentence jumps away from the personalized `{{First Line}}`]

### Priority Improvements
1. [Most important fix]
2. [Second most important]
3. [Third]

Task 7: Score Decision

Hard gate (overrides the score): if ANY rule marked HARD or REJECT above is violated - a static body over 80 words (outside the one value-stack exception), a missing or weak offer CTA (including Email 1), a salesy follow-up ending ("Last note from me", "just following up", etc.), a [bracket] AI variable instead of {{...}}, {sender_signature}, a missing/malformed sender-name signature block, a NEVER-tier spam word, a blown spam budget (2+ HIGH or 3+ MEDIUM in one email, or any tiered word/$/% in a subject line), an Email 1 that fails the one-breath offer test, any AI-tell word or structure from sound-human.md, a failed list test or favor test (marketing-email / brochure framing), or a missing ICP Voice Card - the review fails regardless of the numeric score. Fix every HARD/REJECT violation (mechanically where possible: variable syntax, signature swap, trimming to the ceiling, swapping in the safe alternative from spam-words.md) and re-check before writing emails_v2.md.

If score >= 4.5 AND no HARD/REJECT violation remains:

• Apply all identified fixes to the email copy
• Write the corrected version as {segment_folder}/emails_v2.md
• Report success: "Email review passed with score [X]. emails_v2.md written."

If score < 4.5 OR any HARD/REJECT violation remains:

• Do NOT write emails_v2.md
• Output the full review with specific revision instructions
• Signal that email-copywriting needs to revise:

  Email review score: [X]/5 - Below 4.5 threshold.
  emails_v2.md NOT created. Email copy needs revision.
  
  Revision instructions for email-copywriting:
  [Specific, actionable fixes organized by email]
  

Review Priority Rules

When encountering conflicting guidance, apply in this order:

1. Client notes override everything - If notes.md says "client hates emojis", never use emojis even if other guidance suggests it
2. Conversation feedback overrides notes.md - If user said something in this session, it's the latest preference
3. Campaign strategy overrides generic guidelines - The strategy was created for this specific campaign
4. QA checklist provides baseline - Use as default when no specific guidance exists

Review Scope

ONLY review:

• Subject lines
• Email body content
• CTAs
• Formatting and readability
• Tone alignment with client preferences

DO NOT review:

• AI personalization prompts (handled by prompt-writer skill)
• List building criteria
• Campaign strategy itself

Output Handling

If segment folder specified (workflow mode):

1. Write review output to conversation
2. If score >= 4.5: Write emails_v2.md to segment folder
3. Report result to calling skill/workflow

If standalone mode:

1. Output review to conversation
2. If score >= 4.5: Write emails_v2.md to segment folder (if folder known) or return inline if no folder is known

Resources

For detailed review criteria, see the following reference files in this skill folder:

• qa.md - QA checklists, scoring criteria, self-review questions
• extra_info.md - Quality exercises, writing tips, common mistakes to avoid