Toss Converts Log Definition Work into AI Skills to Ensure Consistency

• Converted the manual log definition work of DataOps Managers into AI skills to automatically reflect domain context and rules
• Established a role division structure where the skill proposes reuse of existing specifications, lctl performs format checks, and Log Center controls final issuance
• Separated documentation structure by task and integrated deployment into lctl to extend from individual tools to team standards
• Confirmed during evaluation that domain context, rather than rules, determines definition quality, with plans to improve accuracy through future knowledge base accumulation

How AI Skills Work

• The skill queries screen designs and existing log specifications to propose reuse candidates and requests human confirmation for uncertain parts
• lctl checks mechanically determinable format rules before saving, designed to follow the latest standards even when rules change
• Log Center manages final quality by rejecting the issuance of specifications that do not meet the rules

Collaboration and Improvement Process

• Collaborated with the Log Product Team to include the skill in lctl for deployment and improved the documentation structure to prevent AI rule omissions
• Previously, issues were discovered in the field and modifications were requested, but this has shifted to a model where DOM directly develops using AI skills and the platform organization handles integrated deployment
• Confirmed through peer feedback that it is now possible to prevent duplicate issuance and maintain connectivity with existing logs