prompt-writer

Prompt Writer Skill

Generate detailed AI personalization prompts based on campaign strategy AI variables. These prompts instruct your personalization tool or the LLM to create personalized content for each prospect.

Output Contract

prompts.md is the handoff artifact for personalization. Hard rules this skill must honor:

• ## User Prompts (To output: true) H2 heading - exact phrasing.
• One H3 per variable using Title Case (### First Line, not ### first_line). H3 name is ≤30 chars.
• Every H3 must contain a **Prompt**: marker followed by the prompt text.
• Prompt text ≤4096 chars. Runs until next ### , ## , or ---. Do NOT use --- inside a prompt body - it cuts the prompt short.
• Variable names must match exactly what emails_v2.md uses in {{Title Case}} braces (e.g. {{First Line}}, {{PS Line}}, {{Subject Line}}).
• Optional ## Rewriting Instructions H2 at the bottom - a short steering paragraph for a rewriter pass, if you use one.

Plan Folder Resolution

This skill receives a {segment_folder} path (e.g., {workspace}/03-15 - plan 1/1 - early-stage/). Resolve {plan_folder} as the parent directory of {segment_folder} when you need plan-level files such as plan.md.

Execution Tasks

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

Task 1: Load Context

• Read {segment_folder}/strategy.md - Extract AI variables section
• Read {segment_folder}/emails_v2.md - Reviewed email copy (fall back to emails.md if not found)
• Read {workspace}/research.md - ICP, value prop, messaging themes, product info, case studies ({workspace} = the client campaign workspace, default ./cold-email/{slug}/ under the current directory)
• Read {workspace}/notes.md (if exists) - Client-specific preferences

Task 2: Build User Prompts

Note: You write user prompts that generate personalized variables (to_output: true). Company context, base rules, and the email sequence are supporting context for the LLM - pull them from research and the sequence file rather than inventing a separate backend. The default rewriting rules (core principles, style, spam/deliverability, flow) are tool-side if your stack rewrites outputs - you SHOULD still write a short optional steering paragraph for rewriting (see Task 2b). The to_rewrite field on each user prompt controls whether that variable's output should go through a rewriter.