pi-code-review

A Pi extension that provides automated, language-aware code review after the agent writes or modifies files.

Package details

← Back
extension

Install pi-code-review from npm and Pi will load the resources declared by the package manifest.

npm repohome report
$ pi install npm:pi-code-review
Package
pi-code-review
Version
0.2.0
Published
Apr 11, 2026
Downloads
141/mo · 6/wk
Author
mdevy
License
MIT
Types
extension
Size
63.1 KB
Dependencies
0 dependencies · 4 peers
Pi manifest JSON
{
  "extensions": [
    "dist/index.js"
  ]
}

Security note

Pi packages can execute code and influence agent behavior. Review the source before installing third-party packages.

README

pi-code-review

A Pi extension that provides automated, language-aware code review after the agent writes or modifies files.

Installation

pi install npm:pi-code-review

Features

Automatic review (zero cost)

After each turn where the agent edits files, a language-aware review checklist is injected into the system prompt. The agent self-reviews before proceeding, catching type safety issues, error handling gaps, security concerns, and naming problems.

Supports: TypeScript, Python, Go, Rust, Java, PHP.

On-demand review (/review)

Run a thorough code review with structured findings:

/review                    # review all uncommitted changes
/review --staged           # only staged changes
/review --ref=main         # diff against main
/review src/foo.ts         # specific files

When an Anthropic API key is available, /review uses a direct Haiku call for structured output with severity-leveled findings (CRITICAL / HIGH / MEDIUM / INFO). Without an API key, it falls back to a prompt-based review via the session agent.

How it works

  1. Edit tracking: hooks into tool_execution_end to collect files modified by Write/Edit tools during each turn
  2. Turn batching: at turn_end, snapshots the accumulated edits (no per-edit overhead)
  3. Prompt injection: at before_agent_start, injects a brief language-specific review checklist into the system prompt
  4. On-demand: /review reads file contents, calls Haiku for structured analysis, and formats findings with severity, line numbers, and suggestions

License

MIT