agent-run-forensics

This plugin reconstructs and audits past Claude Code and Codex agent runs by analyzing session transcripts recorded on disk, providing a finder for locating runs by criteria, a timeline that traces every step with exit codes and file changes, and a claim checker that validates summaries against actual recorded events. It uses three read-only Python scripts to answer questions like "why did the agent do that?" and catch discrepancies such as tests passing before later edits, without making model calls or accessing external networks.