conversation-format

To parse Claude Code JSONL conversation logs, this Skill extracts user and assistant messages while intelligently filtering out noise entries like progress updates and internal system metadata. It differentiates content blocks such as text, thinking, and tool uses to isolate the core narrative.

Filtering noise from LLM logs is achieved by discarding irrelevant entries such as internal system events and progress updates. The Skill identifies valuable messages and crucial metadata like turn duration, ensuring you only process essential narrative elements from the conversation data.

You can scan Claude Code conversation files for PII and secrets using the built-in PII scanning mechanism. This feature flags potential Personally Identifiable Information found within the extracted conversation data, helping maintain security compliance during analysis.

Linking subagent transcripts to tool uses in Claude Code conversation data is supported through a dedicated subagent linking method. This feature connects main conversation tool uses to their corresponding subagent transcripts, allowing comprehensive tracking of AI decision-making.

This conversation parsing approach selectively handles embedded images in JSONL files. It extracts content blocks such as text, thinking, and tool uses, with special handling specifically applied to parse and process any embedded image data found within the conversation logs.