| name | user-guide |
| description | Comprehensive guide to prx features, shortcuts, and configuration. Use when the user asks about prx itself — how to use it, what shortcuts are available, or how to configure scoring. |
prx User Guide
prx is a conversation-first terminal UI for prioritizing code review. It uses AI to score each PR on how much human judgment it requires.
How It Works
- prx fetches your open PRs and scores each one on configurable risk criteria
- You see the assessment as scrollback content in a chat interface
- You interact via chat messages, slash commands, or keyboard shortcuts
- A diff overlay and bulk approve screen are available as secondary views
Scenes
- Conversation (primary): Chat + PR assessment. Type messages or slash commands.
- Diff Overlay (
ctrl+d or /diff): Full-screen diff with file/hunk navigation and inline comments.
- Bulk Approve (
/bulk): Select and approve multiple low-risk PRs at once.
Common Questions
How do I navigate between PRs?
ctrl+n next, ctrl+p previous. Or type /next and /prev.
How do I see the diff?
ctrl+d opens the diff overlay. q or esc returns to conversation. On return visits, the diff automatically shows only what's new — unchanged hunks are collapsed.
What is incremental review?
prx tracks what you've seen by snapshotting hunk and comment content hashes when you exit the diff. On your next visit (after a refresh or app restart), if anything changed, the diff opens in incremental mode: new hunks are expanded with a green "new" badge, unchanged hunks collapse, and new comments get a green badge. Edited comments get a blue "edited" badge. The scoring panel shows a green "New" summary line. Use > to expand everything back to the full diff, or {/} to jump between only the expanded (new) hunks.
How do I approve/merge/reject?
Type /approve, /merge, or /reject. Or just ask in chat: "approve this PR". All actions require confirmation.
How do I add a comment?
Type /comment to open the comment modal in the diff view. Or press c while in the diff overlay to comment on the current line.
How do I change the scoring model?
Ask in chat: "change the model to opus". Or use the set_model config tool.
How do I adjust what gets auto-approved?
Ask in chat: "set approve threshold to 2.5" or "set review threshold to 4.0". Config file: ~/.config/prx/config.toml under [thresholds].
How do I customize scoring criteria?
Ask in chat: "add a criterion for security" or "remove the novelty criterion" or "increase blast radius weight to 2.0". Criteria live in ~/.config/prx/config.toml under [[criteria]].
How do I change the merge method?
Ask in chat: "change merge method to squash". Options: merge, squash, rebase.
What do the verdicts mean?
- APPROVE: score below
approve_below (default 2.0) — safe to approve quickly
- REVIEW: between thresholds — needs normal review
- INVESTIGATE: score above
review_above (default 3.5) — needs careful attention
How do I add custom skills?
Create a directory in ~/.config/prx/skills/ with a SKILL.md file. See the reference guide for the format.
Full Reference
For the complete keyboard shortcut tables, all slash commands, scoring criteria details, configuration format, and custom skills documentation, read the bundled resource reference/guide.md.