/pyor:review – Claude skill for human code review

Agent tooling for Pyor, the native home for GitHub code review. Works with Claude Code, Codex, Cursor, and any coding agent.

Watch the 5-minute demo: /pyor:review in Claude Code, notes in Pyor, fixes back in the branch.

Open your working changes as a local pre-PR review in the Pyor desktop app, before a PR exists. /pyor:review has a panel of fresh sub-agents read the diff and hand Pyor two things: file groups (the changes sorted into labelled folders, important first) and inline hints (short pointers to risks worth a look). No API key, no in-app model run.

The AI here is deliberately quiet: it points you at what matters and gets out of the way. No summaries, no walls of generated text to wade through, just groups to orient you and the occasional hint on a line worth a second look. You read the actual code.

/pyor:review , the primary command.Authors an AI review by default. A panel offresh sub-agents produces the file grouping + inline hints from the diff, then hands them to Pyor to render. The reviewers run withno context from the calling session (they see only the code, so the review isn't biased by whoever wrote it) across three lenses: correctness, simplicity, and security. After you read and comment in Pyor, click**"Send to Claude"** to push your comments straight back into the session that opened it.Takes an optional grouping intent:

/pyor:review                      # importance: signal vs noise, important first (default)
/pyor:review walkthrough          # order the groups as a reading path through the change
/pyor:review custom "by feature"  # your own grouping instruction
/pyor:review plain                # skip the AI panel, just open the diff (alias: fast)

Both open your working changes as a local pre-PR review, ready to read before a PR exists; re-running reopens the same review (idempotent).

Claude Code, install the plugin:

/plugin marketplace add Pyor-review/pyor-cli
/plugin install pyor@pyor

Any other agent (Codex, Cursor, Amp, Gemini CLI, Windsurf, Zed, …), install the skill with skills, which mirrors it into your agent's format:

npx skills add https://github.com/Pyor-review/pyor-cli/tree/main/plugins/pyor/skills/pyor-review

Either way, then run /pyor:review (or invoke the pyor-review skill) from any repository. Both drive the same pyor-review flow.

• The Pyor desktop app . It registers thepyor:// URL scheme, renders the review, and exports the review context the AI panel writes against. Grab it atpyor.review ;/pyor:review hands you the install command if it's missing.
• git andnode on your PATH.
• gh (the GitHub CLI) is optional, used only when a command needs GitHub data.

Both the Claude plugin and the cross-agent skill drive one small, agent-neutral CLI that does the deterministic parts: git resolution, the working-tree revision token (which must match what the app computes so aids cache-hit), the pyor:// deep link, and the ~/.pyor feedback channel. The agent only supplies the analysis (grouping + hints).

Run it without installing:

npx -y pyor-review prepare
npx -y pyor-review --selftest

Subcommands: prepare (resolve repo/head/base + a deterministic session id + revision + the exported review context), open (fire the deep link with the aids), wait (park for "Send to Claude" feedback). pyor-local-review is the plain, no-AI open.

The scripts are self-contained and repo-agnostic, sharing plugins/pyor/scripts/lib.mjs:

node plugins/pyor/scripts/pyor-local-review.mjs --print     # print the deep link, do not open
node plugins/pyor/scripts/pyor-ai-review.mjs prepare        # resolve repo/head/base + revision + context
node plugins/pyor/scripts/pyor-ai-review.mjs --selftest     # run the built-in assertions (incl. session-id determinism)