opencode-conversation-analysis

Analyze OpenCode conversation history by processing chunked JSONL outputs from stored messages to identify recurring themes, steering patterns, and decision-making styles across multiple sessions. The analysis produces a structured JSON with theme counts and representative quotes for ranking.

Agent steering in AI coding assistant conversations refers to how users guide the assistant's behavior through specific communication patterns and cues. Analyzing these patterns from chunked JSONL session data surfaces user intents, guidance styles, and decision-making trends for product optimization.

Yes, you can extract conversation themes across multiple sessions and timeframes. The analysis applies parallel subagent processing to chunked JSONL outputs, aggregating chunk-level insights into a global theme report that spans the entire conversation history.

The conversation analysis outputs a structured JSON format containing identified themes, occurrence counts, and representative quotes. This quantitative theme ranking is suitable for product optimization and developer tooling display.

Prepare JSONL files by running the extraction workflow to generate chunked data from OpenCode conversation history. These extracted chunks serve as the input for the analysis tool to derive themes, user-guidance patterns, and decision-making styles.

Parallel subagent analysis provides scalable processing for large JSONL datasets by yielding chunk-level insights that aggregate into a global theme report. Without this approach, processing large conversation histories across multiple sessions would lack the necessary scalability.