Skip to content

PFigs/redliner

Repository files navigation

redliner

A local review tool for AI agent work. Comment on plans and code diffs line-by-line, resolve feedback, and approve changes through a web UI or CLI.

The idea: an agent writes code, you redline it. Comments are stored as JSON files that the agent can read and act on.

Install

uv tool install redliner

Plan review

Review a plan file with inline comments:

redliner open plan.md        # web UI in your browser
redliner show plan.md        # print with line numbers and comments
redliner comment plan.md 10 "This needs more detail"
redliner list plan.md        # list all comments
redliner resolve plan.md 1   # resolve comment #1
redliner approve plan.md     # approve (fails if unresolved comments)

Diff review

Review working tree changes side by side:

redliner diff                 # opens all changes in web UI
redliner diff --path src/foo.py   # single file
redliner diff --no-open       # just print status as JSON

The diff view shows old and new versions in collapsible columns. Click any line to leave a comment.

How comments are stored

Comments are stored under $XDG_DATA_HOME/redliner/reviews/ (defaults to ~/.local/share/redliner/reviews/). Each review session (a plan file, or a diff over a repo) gets its own directory with comments.jsonl and meta.json.

comments.jsonl has one comment per line, each self-describing with the full file path:

{"id": 1, "file": "/abs/path/plan.md", "line": 42, "text": "Handle the None case", "status": "pending", "created": "..."}

meta.json tracks per-file review status:

{"files": {"/abs/path/plan.md": {"status": "approved", "approved_at": "..."}}}

The storage path is shown in the web UI header with a one-click copy button. From the CLI, redliner status <file> prints it as "storage". Agents can read comments.jsonl directly.

Agent workflow

  1. Agent writes code or a plan
  2. You run redliner diff or redliner open plan.md
  3. You leave comments on specific lines
  4. Agent reads the review JSON files and addresses feedback
  5. You review again until satisfied, then approve

Requirements

Python 3.12+. No external dependencies.

About

A python tool to collaborate with agents when reviewing and editing a plan.

Resources

License

Stars

Watchers

Forks

Packages

 
 
 

Contributors

Languages