User repository on GitHub
Isolated AI-powered code and plan reviews from the command line
- Forks
- 1
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.Homepage
CommunityJev Agents & MCP
Pi Review
Isolated multi-reviewer code review as CLI, CI gate, and agent skill, running fresh read-only Pi sessions singly or as a consensus panel with structured verdicts and exit codes.
- Stars
- 15
- Language
- TypeScript
- License
- MIT
- Published by
- Community
- Author
- ZephyrDeng
- Last verified
- Oct 2026
About this project
Isolated, multi-reviewer AI code review — as a CLI, a CI gate, and an agent skill. Quick start · Features · Screening workflow · Panel review · CLI reference · 中文说明. pi-review hands review work to a fresh, read-only Pi session and returns a structured verdict: findings with severity, evidence, location and a stable exit code. Run one reviewer, or a panel of independent reviewers that must agree before a finding blocks the gate. Your agent (Claude Code, Codex, Cursor, Pi) stays the only editor.
Quoted from the project's README. Read the full README.
Works forJev for Classification
Used inJev for Software & DevTools
- Languages
- TypeScript
Highlights
- Fresh read-only reviewer processes with allowlisted tools plus machine-readable JSON lines and stable exit codes.
- Panels of two to eight independent reviewers with any, quorum, majority, or unanimous blocking thresholds.
- Fast screening path combining deterministic slicing with typed Jev judgments against a growing defect catalog.
- Bounded fix and re-review loops with cross-round diffing plus live terminal, web dashboard, and CI hooks.
Watch out
Reviewers are read-only and failures never silently pass: child failure is blocked and unparseable output needs a human. Panel clustering only groups findings and cannot invent, drop, or rewrite them. Copy reviewed against the repository README on 2026-10-04, not from running it.
Questions about this entry
Short answers drawn from the linked project or page.
No. Every review spawns a fresh child with a read, grep, find, and ls allowlist that cannot edit or commit. Your agent stays the only editor while reviewers only return findings.
Use screen for a fast deterministic plus Jev catalog pass when you want a quick gate, and full or panel review when you need deep correctness, security, and test lenses. Screen is far faster but narrower.
Reviewers never see each other, deterministic matching clusters by path and summary, and ambiguous pairs get Jev adjudication with borderline re-judging. Findings block only when enough reviewers agree.
They are optional in different places. pi-review itself needs Node 20 or newer plus the Pi CLI with at least one model provider configured. TYPESAFE_API_KEY adds Jev consensus adjudication, cross-round finding memory, scope classify, and the roughly 1.2 second screen gate; without it the reviewers, panels, and loop all still run, but nothing is silently skipped into a pass. A failed reviewer exits blocked and unparseable output exits needs_human, with exit codes 0, 1, 3, and 4.
npm install -g @zephyrdeng/pi-review for the CLI, or npx @zephyrdeng/pi-review install for the Pi package plus a skill for Claude Code, Codex, Cursor or Antigravity. It needs Node 20 or newer and the Pi CLI with at least one model provider configured. In Pi you get the /rv, /rv-loop and /rv-models commands; other hosts ask for a review in plain language.
Yes, and that is the point of cross-family panels: --reviewer-model r1=openai/gpt-5.6:high --reviewer-model r2=anthropic/claude-opus-4.8:xhigh. Panels hold 2 to 8 independent reviewers who never see each other, and a finding blocks only when enough agree under any, quorum, majority or unanimous. Adjudication only clusters findings and cannot invent, drop or rewrite them.
Reviewers read .claude/rules/ with Claude Code semantics, appending rules without a paths key to the system prompt and injecting path-scoped rules only when a read matches; write and edit never trigger them, and --no-rules or PI_REVIEW_RULES=0 disables them for a run. Separately, --mode selects code lenses, a six-lens plan review, or a challenge adversarial pass.
More Jev resources like this
48GitHub stars
Agent harness for work that outlives one chat, with resumable turns, routed state machines, and observational memory across sessions and processes.
Jev Agents & MCP#community#jevbest#catalog-import
Community
dzhng/duet-agentTypeScript
dzhng/duet-agentTypeScript37GitHub stars
Assistant framework that stores specs, memory, and specialist agents as plain markdown files in your repo for Claude Code, Codex, Opencode, and similar runtimes.
Jev Agents & MCP#community#jevbest#catalog-import
36GitHub stars
Obsidian plugin providing multi-provider chat, vault read and write with function calling, visual workflow automation, and local vector search.
Jev Agents & MCP#community#jevbest#catalog-import
Community
takeshy/obsidian-llm-hubTypeScript
takeshy/obsidian-llm-hubTypeScript36GitHub stars
Pi extension that strips the full Agent Skills catalog from every request and loads only the skills a Jev ranking call judges relevant.
Jev Agents & MCP#community#jevbest#catalog-import
102GitHub stars
Behavioral code search that asks Jev to score repository fragments for a plain-language question and returns exact excerpts with paths and lines.
Jev Agents & MCP#community#jevbest#catalog-import
Community
nassim-arifette/jevgrepTypeScript
nassim-arifette/jevgrepTypeScript44GitHub stars
Pi extension that keeps a fast Executor model on implementation and brings in a separate Advisor model for consequential choices and reviews.
Jev Agents & MCP#community#jevbest#catalog-import
Community
philipbrembeck/pi-advisorTypeScript
philipbrembeck/pi-advisorTypeScriptFrom the Jev community
Posts from builders shipping with Jev right now.