scan-abandoned-sessions

To find abandoned Claude Code sessions, you can scan JSONL session log tails against progress heuristics. This filters out completed tasks and noise, reporting only unfinished work and the last user prompt from your project directories.

Recovering unfinished Claude Code work involves auditing session logs with progress heuristics to identify interruptions. The scan classifies abandoned sessions contextually, summarizing the exact point of interruption so you can resume quickly.

The session log audit distinguishes abandoned tasks by analyzing JSONL file tails against predefined progress heuristics. It operates on project-specific directories to automatically filter out noise, completed tasks, git-flow pauses, and ongoing work.

Yes, the abandoned session scan requires file system access to read JSONL session logs and maintain a scan cursor. This access enables incremental auditing of project-specific session directories without reprocessing previously checked data.

Yes, you can audit session logs incrementally because the scan maintains a cursor within the file system. This cursor tracks previously processed JSONL data, ensuring the scan only evaluates recent session log tails for abandoned work.

Completed tasks or git pauses may appear if session log tails are not properly filtered. The scan uses intelligent filtering and predefined progress heuristics to automatically classify and exclude these non-abandoned states from the final report.