/explore

Start an interactive transcript exploration session. Present initial findings from the transcript corpus, then let the user steer deeper investigation.

Arguments

• --project <name> — Scope to a specific project transcript directory. Default: current project.

Execution Steps

1. Resolve transcript path. Same logic as /agentskill-kaizen:analyze — derive project key from --project flag or current working directory.

2. Run initial survey. Use DuckDB MCP to run a quick corpus overview:

   • Total session count and date range
   • Record type distribution
   • Top 10 most-used tools
   • Error rate summary
   • User interrupt count

3. Present findings to user. Display the survey results and suggest investigation directions:

   • "I found {N} sessions with {M} tool misuse violations. Want to dig into those?"
   • "There are {K} user corrections. Want to see the frustration signals?"
   • "The most common error is {type} ({count} times). Want to trace those sessions?"

4. Follow user direction. Based on user response, run targeted queries:

   • Use DuckDB SQL for structured data extraction
   • Use kaizen MCP tools for process mining and pattern detection
   • Present results incrementally
   • Ask clarifying questions to narrow investigation

5. Save findings on request. When the user wants to save findings, write to .planning/kaizen/exploration-{date}.md.

Interaction Pattern

This command runs interactively — do NOT spawn an autonomous agent. Stay in the main conversation so the user can steer the investigation in real-time. Use MCP tools directly for queries.