handoverlisted

# handover Compaction throws history away. The raw `.jsonl` transcripts keep all of it, but they are unreadable and enormous. This builds a small greppable pack from them so you can `/clear` instead of `/compact` and pull back only what you need. ## Build or refresh the pack ``` python ~/.claude/skills/handover/scripts/build_context.py ``` Run it from the project root. It finds that project's transcripts under `~/.claude/projects/<slugified-cwd>/` on its own and writes `.handover/` into the project. If `python` is not on PATH, use `py -3`. Flags, all optional: - `--project <dir>` project root, defaults to the current directory - `--in <dir>` transcript directory, when auto-discovery cannot find it - `--out <dir>` pack location, defaults to `<project>/.handover` - `--rebuild` discard the pack and import everything from scratch Reruns are cheap and safe. A manifest at `.handover/.state.json` records a byte offset per transcript, so a rerun reads only new bytes. The turn in flight when you run it is held back and rendered whole on the next run. ## What the pack contains - `TIMELINE.md` — every turn from every session, merged into one chronological stream. The entry point. Read this first on a cold start. - `INDEX.tsv` — one row per prompt, reply, file write, command, error and decision. Every row ends in a `sessions/<file>.md:<line>` jump target. - `FILES.tsv` — each project file, how often it was written, when, and by which sessions. - `sessions/<date>-<time>-<id