alingse/human-review

Overview

This skill lets AI agents launch a web-based human code review session, wait for reviewer comments, parse the results, and apply suggested changes automatically. It integrates with the hrevu CLI to manage the review lifecycle and then uses programmatic edits to implement actionable feedback. The goal is to combine human judgement with automated edits while preserving code formatting and context.

How this skill works

The skill runs the hrevu CLI against a file, commit, or current git diff to open a browser-based review. It blocks until the reviewer finishes and hrevu emits a structured summary of comments to the terminal. The skill parses each comment for file, line, context, and suggested change, filters out non-actionable praise, and maps actionable items to edit operations. Edits are applied with the Edit tool, using replace, replace_all, or insert operations as appropriate, and each change is reported back.

When to use it

• When you want explicit human feedback on code before applying automated changes.
• When reviewers will provide line-level suggestions, renames, or code additions through a web interface.
• When you prefer a controlled flow: open review → wait → parse → apply edits.
• When maintaining formatting and context is important during automated modifications.

Best practices

• Install and test the hrevu CLI (cargo install human-review) before using the skill.
• Launch hrevu with a specific file or commit to limit scope and reduce ambiguity.
• Skip comments that are pure praise or vague; confirm with the reviewer if the change is unclear.
• Preserve original indentation and formatting when applying edits; report location and content after each change.
• Use replace_all for systematic renames and targeted edits for localized fixes.
• Always wait for hrevu to finish and emit the summary before parsing and applying changes.

Example use cases

• Open a review for a single file with hrevu <file>, wait for reviewer suggestions, then apply minor fixes and import additions.
• Run hrevu on a commit to gather multi-file feedback and apply explicit code changes suggested by the reviewer.
• Use hrevu diff when there are unstaged changes to get focused human input on recent edits before committing.
• Handle variable or function renames by applying replace_all after confirmation from review comments.
• Add recommended error handling or logging lines suggested in review comments, preserving surrounding context.