Plugin Review

This page is auto-generated from the plugin's README. Do not edit directly — changes will be overwritten on the next sync.

Plugin Review

Plugin quality review covering principles alignment, terminal UX, and documentation freshness via orchestrated subagents.

Summary

plugin-review runs a multi-pass, assertion-driven audit of any Claude Code plugin. Four read-only analyst subagents work in parallel across four tracks (principles alignment, terminal UX quality, documentation freshness, and context efficiency) then report findings back to the orchestrator, which auto-implements all fixes and re-audits until the assertion confidence score reaches 100% or the pass budget is exhausted.

An optional --autonomous flag adds a fifth subagent (regression guard), tier-classified auto-fixing, and build/test validation after each implementation pass.

Principles

[P1] Act on Intent: Invoking /review is consent to the full multi-pass loop. No approval gates during the loop.

[P2] Scope Fidelity: The orchestrator does not read full plugin source files. Subagents analyze; the orchestrator implements.

[P3] On-Demand Template Loading: Templates are loaded by the component that uses them, not pre-loaded.

[P4] Bounded User Interaction: Every decision point uses AskUserQuestion. During the review loop itself no interaction occurs.

[P5] Convergence is the Contract: The loop drives toward 100% assertion confidence without check-ins.

[P6] Documentation Co-mutation: Every implementation change must include corresponding documentation updates.

[P7] Analyst/Orchestrator Separation: Read-only analysts (produce findings only) are strictly separated from write-capable fixers.

[P8] Enforcement Layers: Mechanical > Structural > Behavioral.

[P9] Read-Only Analyst Enforcement: A PostToolUse hook blocks when write tools are added to analyst agent YAML.

Installation

/plugin marketplace add L3DigitalNet/Claude-Code-Plugins
/plugin install plugin-review@l3digitalnet-plugins