Langfuse Feedback Analyst Skill

Purpose

Analyzes user feedback from Langfuse annotation queues and generates surgical recommendations for template.yaml, style.yaml, and tools.yaml configuration files.

Architecture Context:

• template.yaml: Defines document structure requirements (sections, fields, content organization)
• style.yaml: 5-layer style schema (constraints, signatures, patterns, dynamics) compiled into 30+ validation checks
• tools.yaml: Research tool configuration (patterns, atomic tool selection, composite orchestration)

The skill maps feedback patterns to specific configuration layers, producing targeted YAML edits that address quality issues found in production traces.

CRITICAL CONSTRAINT: Only suggest changes to config fields that are actually read and used by the generation architecture. Consult CONFIG_FIELD_USAGE_MAP.md before making recommendations - it contains the definitive list of fields that affect generation. See learnings/pitfalls.md for common mistakes to avoid.

Required Inputs

1. Annotation Queue ID: Langfuse queue containing feedback (scores and comments if applicable)
2. Case Folder: Path to case config directory or name of the case to look at (all cases are in the writing ecosystem config cases)