Read changes with context
Inspect commits and branch comparisons without losing sight of the surrounding behavior.
A diff tells you where to begin a review. Understanding the change usually requires reading the code around it as well.
Choose the comparison
Use the Explorer’s Git comparison view to inspect a commit or compare branches. A single-commit comparison uses its first parent. Branch comparisons can use a direct comparison or a merge-base comparison, depending on the question you want to answer.
Confirm the base, head, and comparison direction before reviewing. Comparing the same two branches in a different direction changes the meaning of additions and deletions.
Yenpo fixes the selected comparison to commit identities for that review, so the scope does not silently follow a branch as it moves.
Open the changed files
Explore the changed-file tree and open a file in its read-only inline diff tab. Historical files can remain open alongside the current workspace version of the same path.
Keep track of which version you are reading. The behavior in a historical comparison may differ from your current working tree.
Investigate the behavior
Add the comparison to an AI Reader conversation and state the question you want to investigate:
Review this change for cancellation and error-handling regressions. Explain each finding’s trigger and show the supporting code.
A useful finding should identify a location, a concrete trigger, an observable consequence, and the evidence connecting them. A general preference about style is a different kind of observation.
Read relevant callers and tests before accepting a conclusion. A line that looks unsafe in isolation may be protected by a caller’s contract; a safe-looking helper may become unsafe when a new caller violates that contract.
Keep review and editing separate
Fixed Git comparisons are always read-only. If you decide a fix is needed, start a separate task against the current working tree and explicitly authorize the edit for that run.
After any edit, inspect the resulting change and run the checks appropriate to the affected behavior. An AI explanation is not a replacement for verifying the result.