What problem does it solve? Writing personalized cold email lines for hundreds of leads by hand is slow and inconsistent, and running a single large prompt over a full lead list produces generic, low-quality output. This Skill defines a reusable approval-loop pattern that tunes a personalization prompt on small batches before scaling to the full list. ## Core Features & Use Cases - Approval loop before scaling: Shows one sample personalization, then batches of 10, collecting user edits until two consecutive rounds pass with zero corrections before locking the prompt. - Parallel Task sub-agent fan-out: Splits leads into batches of 10-20 and launches multiple Claude Code Task sub-agents (including A/B/C copy variants) that write JSON results to /tmp for merging by lead_id. - Schema and quality guardrails: Defines a strict output schema (situation_line, value_line, cta_soft), banned-word rules, no-fabrication rules, retry handling for malformed JSON, and skip flags for leads with thin data. - Use Case: Given 100 enriched leads with company descriptions, tune the prompt over two approval rounds, then fan out 10 parallel sub-agents to produce three personalization variants per lead, merged and mapped to Smartlead custom variables like {{situation_line_a}}. ## Quick Start Use the personalization sub-agent pattern to generate situation_line, value_line, and cta_soft fields for my lead list, starting with one sample lead for my approval before fanning out in parallel batches.
