NAME

    Review::Helper - Web app for reviewing changes written by AI agents.

EARLY VERSION

    This is a very early version. Some key features are still to come, and
    it is not ready for daily use. Expect changes between releases,
    including to the commands, the review page and the review file format;
    a review built with one release may not open in the next.

DESCRIPTION

    Review::Helper helps a human review code an AI agent wrote. When the
    agent finishes a change (a commit, a branch, a pull request) it builds
    a review file describing the change and fills in explanations: what,
    why and how for the whole change and for each commit, file and sub. The
    human opens the review in a local web app, reads the change top-down,
    leaves feedback, and exports it back to the agent, which acts on it and
    adds a new iteration.

    While writing code, the agent records notes about wrong assumptions,
    deviations from a plan, and anything else the reviewer should see.
    Those that need the reviewer's judgment become flags beside the lines
    they concern: yellow when a reviewer may object, red when something is
    probably still wrong. The rest are listed in the iteration's journal.

    The other primary use is project review: reviewing a whole codebase, or
    one area of it, as it finally lands. With no notes recorded, the agent
    deduces points of interest from the commit history, docs, tests, and
    issues, and builds one combined review of the final code with a guided
    tour, rather than going commit by commit. A large codebase can be split
    into several reviews by path. This also serves for learning a codebase.

    The review file is a SQLite database kept inside the repository under
    review, by default in the worktree's git directory, so it never dirties
    a working tree. It stores only review data. Diffs and file contents are
    read live from git, and each iteration's head commit is pinned under
    refs/review-helper/ so rebases and garbage collection do not remove it.

    Everything is driven by the review-helper command. Agents never edit
    the review file or the refs directly; they use its subcommands.

    One goal of this project is a standard format for AI-assisted code
    review: a sharable review file plus the tools and agent skills around
    it. The format, specified in standards/FORMAT.md of the share directory
    (see "THE REVIEW FILE FORMAT" in Review::Helper::Manual), is documented
    and language-neutral, so other review tools, skills and processes can
    be built on the same file and extend it, with this distribution as one
    starting point.

SYNOPSIS

    Record a note while working:

        review-helper note --flag yellow --kind deviation --path lib/Foo.pm \
            --unit bar --sentence 'Skips expiry when TTL is zero.'

    List the reviews of the current repository, close one, and clean up:

        review-helper list
        review-helper close .git/review-helper/my-change.sqlite
        review-helper prune --dry-run
        review-helper prune

    Report the standard this implementation follows:

        review-helper version

PUBLIC METHODS

    $version = Review::Helper->STANDARD_VERSION

      The version of the Review::Helper standard this distribution
      implements: the review file format, the commands the agent skills
      rely on, and the HTTP API the review page uses. review-helper version
      prints it, and the page's review data carries it.

SOURCE

    The source code repository for Review-Helper can be found at
    https://github.com/exodist/Review-Helper.

MAINTAINERS

    Chad Granum <exodist7@gmail.com>

AUTHORS

    Chad Granum <exodist7@gmail.com>

COPYRIGHT

    Copyright Chad Granum <exodist7@gmail.com>.

    This program is free software; you can redistribute it and/or modify it
    under the same terms as Perl itself.

    See https://dev.perl.org/licenses/