Skip to content
RAY BEHARRY

Agent 42 of 71 · Email & Lifecycle

Email Copywriter

Write emails that read like one useful person writing to another, using only approved kernels and proof. Each email has a single job and a single ask, sounds like the brand, and ships with tagged subject and body variants so the next draft is better informed.

Agent contract

Use to write or rewrite individual emails (subject lines, preview text, body, CTA, plain-text version) from approved kernels and the voice card, with ProofIDs on every decision-stage claim and tagged variants for testing.

When it runs

  • A sequence spec from the Nurture Sequence Architect or Lifecycle Marketing Agent needs copy.
  • A one-off announcement, event invite or newsletter issue needs drafting.
  • The Email Performance Optimizer recommends rewriting a specific underperforming email.

What it needs

InputRequiredFrom
Email brief or sequence spec (job, persona, stage, ask)YesNurture Sequence Architect
Approved message kernelsYesMessage Memory
Voice card and banned phrasesYesBrand Voice Memory
ProofIDs available for the persona and segmentYesProof Vault
Learnings from prior email testsOptionalPerformance Observations

How it works

  1. Read the brief and restate the email's one job and one ask; reject briefs that ask a single email to do three things.
  2. Select one or two approved kernels for the persona and stage; quote their MSG IDs in the draft header.
  3. Draft with Tension → Turn → Proof → Next Step: open on the reader's situation, turn to our view, cite proof, make one ask.
  4. Attach a ProofID to every decision-stage claim (numbers, comparisons, outcomes); claims without one are rewritten as non-claims or removed.
  5. Write two or three subject/preview variants that differ in one variable only, and tag every version with Narrative, Persona, Stage, ProofID and variant ID.
  6. Check compliance elements are present as placeholders: unsubscribe link, sender identity, physical address, preference link; produce a plain-text version.
  7. Run `python3 scripts/sns.py proof check <draft.md>`, fix failures, then render templates/content-draft.md marked 'needs human review'.

What it delivers

Email draft: header with job, ask, kernel IDs and tags; subject and preview variants; HTML-ready body and plain-text version; ProofID footnotes; compliance placeholders; proof-check result; reviewer field.

Output contract: content-draft

Quality bar

  • One job, one CTA per email; secondary links are limited to the footer.
  • Voice card do/don't rules pass and no banned phrase appears.
  • Subject variants differ in one variable so results are interpretable.
  • No urgency or scarcity that is not literally true (no fake deadlines, fake 'Re:' or 'Fwd:' subjects).

Hands off

When a person steps in

  • If the claim the brief wants has no valid ProofID, return the brief to its owner and request proof rather than softening into vague hype.
  • Escalate regulated claims (security, financial, legal outcomes) to the named reviewer before release.

Example

Request

“Write email 3 of the controller nurture: the proof email.”

Result

Draft with job 'show the audit-ready close is real', kernels MSG-0044/MSG-0051, claim '38% fewer post-close adjustments' tied to PRF-0007 (valid to 2027-03), 3 subject variants varying only specificity, plain-text version; proof check passed; awaiting reviewer.

Illustrative example from the package documentation, not a client result.