ce-session-inventory

Session metadata extraction scans platform-specific JSONL stores to emit structured fields like platform, ts, session, file, and size. It parses each session format and outputs one JSON object per session, plus a final _meta summary line for pipeline consumption.

Inventorying AI agent session files across repositories requires unified discovery scripts that parse JSONL session stores. This approach extracts per-session data such as timestamps, session IDs, file paths, and platform origin into structured JSONL format for indexing and auditing.

Filtering session logs by repository is supported through the optional --cwd-filter to prune results to a specific working directory. The --keyword option enables keyword scoring mode to prioritize and rank relevant sessions within the structured output.

Extracting session transcripts works across Claude Code, Codex, and Cursor platforms by scanning their specific JSONL stores. The discovery scripts parse each platform's unique format to collect structured metadata including optional fields like branch, cwd, and model.

Extracted session metadata files are output in JSONL format, with one JSON object per session followed by a final _meta summary line. This structured output enables downstream filtering, ranking, and pipeline consumption for auditing purposes.

Parsing AI coding session logs emits structured fields including platform, ts, last_ts, session, file, and size, alongside optional fields like branch, cwd, and model. These fields support comprehensive indexing and auditing of session transcripts across platforms.