code-reviewlisted
Install: claude install-skill seankoji-com/zsh-resume
# Review priorities
Two files carry all the real logic: `bin/resume-sessions` (python3, one
reader per tool that turns an on-disk session store into TSV rows, plus the
repo-scoping filter) and `zsh-resume.plugin.zsh` (flag parsing, the gum /
`select` picker, mapping a picked line back to its row, and the launch).
`spec/fixtures.py` is the third: it is the executable description of what
each tool's store looks like, so a store-format change must land there too.
## Spend real attention here
- Every reader in `bin/resume-sessions` mirrors a format nobody controls. A
diff that changes a field name, a glob, or a skip rule (`thread_source`,
`parent_id`, `/.grok/worktrees/`) needs the matching fixture in
`spec/fixtures.py` changed in the same PR, or the spec is no longer testing
the real shape. Real-world samples are documented in the reader docstrings
and the README table — compare against them.
- Readers must never write. `ro_connect` opens SQLite with `immutable=1` on
purpose (a running CLI holds these files; `mode=ro` fails on WAL stores
and sandboxed dirs). A PR that "simplifies" it to a plain `connect` is a
regression.
- One broken store must not hide the others: each reader is wrapped so it
warns and continues. Check that a new code path keeps that property.
- `RESUME_CMDS` entries are the resume flag per tool (`--resume`,
`--conversation`, `resume`, `--session`, `--resume-id`). These were
verified against each CLI's `--help`; a change needs the new h