← ClaudeAtlas

code-reviewlisted

Review priorities for zsh-resume pull requests — what deserves real scrutiny versus what to skip. Use for every PR review.
seankoji-com/zsh-resume · ★ 0 · Code & Development · score 70
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