[ftabba]

Two messages misstate what a local review reads. The README states that it applies the patches in a temporary scratch clone, and the warning on a dirty working directory states that the reviewer might see the uncommitted changes. Neither happens: run_git_review runs the worker with current_tree set, so nothing is cloned or applied, and the review reads files through git, with HEAD standing for the commit under review.

The first commit corrects the README. The second keeps the warning and its prompt, and makes the warning state that uncommitted changes aren't reviewed.

[sashiko-bot]

Sashiko review

Commit 1/1 — c2255db8 docs: local review reads the checked-out tree

• [MEDIUM] In README.md (Quick Start), the documentation inaccurately states
  that local review reads the checked-out working tree and requires checking
  out the last commit of the range and leaving the working tree unchanged. In
  src/local_review.rs (run_git_review and review_single_patch),
  src/worker/prefetch.rs (prefetch_context), src/toolbox/git_read_files.rs
  (GitReadFilesTool::read_single_file), and src/worker/location_resolver.rs
  (read_file_content), setting current_tree only skips temporary worktree
  creation and patch application while pinning virtual HEAD to each reviewed
  commit SHA and reading committed blobs directly from the Git object
  database rather than the physical working tree.

Full review and stage logs on sashiko.sashiko.dev

[ftabba]

I wrote that the review reads the checked-out tree, but it reads files through git, with HEAD standing for the commit under review. Fixed in the new head, which also corrects the dirty-tree warning: it stated that the review might see uncommitted changes.

[sashiko-bot]

Sashiko review — v2

✓ No issues found across 2 commits.

Full review log on sashiko.sashiko.dev