AI repository intelligence

FR-AI / ANALYSIS

Why it is worth attention

It provides a continuous, session-backed code review workflow that decouples review from model context, allowing the agent to keep working while the user reviews incremental changes with inline comments and two diff modes.

Who it is for

Pi agent users who need to review incremental changes during an interactive session
Developers working with pi in a Git repository who want a persistent review UI
Teams using pi for code generation and requiring a structured review loop without losing context

Use cases

Reviewing and commenting on a batch of changes made by the pi agent before submitting them
Tracking differences between the last review checkpoint and the current workspace to see only new changes
Comparing the full working tree against HEAD while keeping a separate checkpoint for incremental reviews

Strengths

Persistent review window that remains open and updates incrementally as files change
Checkpoint-based review state that survives session saves and restores with pi sessions
Two diff modes (Since review and vs HEAD) and inline comments on both panes
Works with any on-disk change (agent, user, or external tools) without parsing tool calls

Considerations

Requires pi, a Git repository, and Node.js 20+; not a standalone tool
Checkpoint custom entries do not participate in model context, so the AI cannot see review annotations
Review window closes on pi shutdown, session switch, or extension reload, requiring manual re-open

README quick start

Review Loop

A persistent, incremental diff reviewer for pi.

Review Loop keeps a native review window open while the agent works. Submitting a review records the current workspace as a session-backed checkpoint, so the next review can show only the changes made afterward.

Install

pi install git:github.com/earendil-works/pi-review-loop

Update an existing installation with:

pi update --extensions

Start pi in a Git repository and open the reviewer:

/diff-review

The command returns immediately; it does not replace pi's editor. Running /diff-review again brings the existing review window forward.

Review workflow

Open the window with /diff-review.
Select a file from Recently Changed or the Files tree.
Review its diff.
Hover a line number to reveal a +, then click it to add an inline comment.
- The left pane refers to Reviewed content in Since review mode or HEAD in vs HEAD mode.
- The right pane refers to the Current on-disk content.
- Inline comments work on either pane.
Use Add file note for feedback that is not tied to a line.
Comment-count badges appear beside files in both sidebar sections.
Click the review button in the top-right.
- All files currently changed since the checkpoint are marked reviewed, whether or not each file was opened.
- The current workspace becomes the new review checkpoint.
- Review feedback is inserted into pi's normal editor.
- Feedback is not sent automatically; inspect or edit it in pi, then submit it normally.
Keep the window open. Later changes appear as the next review batch.

Submitting with no comments simply marks the current workspace as reviewed.

Diff modes

Since review

Compares:

last review checkpoint → current workspace

This is the default review queue. After submitting a review, it becomes empty until files change again.

Before the first checkpoint, the baseline is the HEAD captured when Review Loop opens.

vs HEAD

Compares:

current HEAD → current workspace

This shows the complete current working-tree change set, including changes already covered by a review checkpoint. Files that still match the last checkpoint display a green reviewed checkmark; files changed afterward retain the blue pending indicator.

Changing modes does no

Description

A persistent incremental diff review loop for pi

Related repositories

Similar projects matched by category, topics, and programming language.

mereyabdenbekuly-ctrl

Featured

clodex-ide

Clodex is an open-source, local-first agentic IDE that combines persistent AI tasks, code, terminal, browser, Git, models, memory, and governed execution in one Electron workspace, currently in technical preview.

AI & Machine LearningAI Agents

859

S40911120

Featured

recensa

Recensa is a self-hosted web viewer that indexes Claude Code session transcripts into a local SQLite database, enabling full-text search, replay, and audit of all past agent conversations without uploading data anywhere.

AI & Machine LearningLarge Language Models

l0ng-ai

Featured

tty7

tty7 is a fast, GPU-accelerated terminal workbench with persistent sessions, built-in input tools, SSH support, and coding agent awareness.

Developer ToolsCLI & Terminal

359