Skill

review-pr

From idh

Multi-perspective code review with parallel agents. Covers correctness, consistency, scope, red team, and doc propagation.

Install

npx claudepluginhub minhhaduong/imperialdragonharness

Tool Access

This skill uses the workspace's default tool permissions.

Preview

Spin multiple agents in parallel, each with a distinct perspective. Run all agents in fresh contexts.

SKILL.md

Similar Skills

kotlin-ktor-patterns

Provides Ktor server patterns for routing DSL, plugins (auth, CORS, serialization), Koin DI, WebSockets, services, and testApplication testing.

everything-claude-code

163.2k

deep-research

Conducts multi-source web research with firecrawl and exa MCPs: searches, scrapes pages, synthesizes cited reports. For deep dives, competitive analysis, tech evaluations, or due diligence.

everything-claude-code

163.2k

inventory-demand-planning

Provides demand forecasting, safety stock optimization, replenishment planning, and promotional lift estimation for multi-location retailers managing 300-800 SKUs.

everything-claude-code

163.2k

Stats

Stars0

Forks0

Last CommitApr 15, 2026

Actions

View Source View Plugin View on GitHub View README

Review PR $ARGUMENTS — multi-perspective agent review

Spin multiple agents in parallel, each with a distinct perspective. Run all agents in fresh contexts.

Setup

Read the issue linked to the PR. Note the exit criteria.

Read the diff of the merge request.

Assess risk level and determine proportional depth (see table below).

Launch review agents in parallel:

Agent

Focus

Key question

Correctness

Logic, edge cases, test coverage

Does this do what the exit criteria say?

Consistency

Style, naming, docs, stale references

Does this fit the rest of the codebase?

Scope

Over-engineering, unrelated changes

Does this change only what the ticket asks?

Red team

Adversarial inputs, broken invariants

How can this break?

Doc propagation

Downstream text accuracy

Do docs and configs still match the code?

Proportional depth

PR risk

Agents

Trivial + user present

Skip PR — merge directly

Trivial (typo, config)

Correctness only

Standard

Correctness + Consistency

Standard + scripts

+ Doc propagation

Substantial

All five

High-risk (schema, methodology)

All five + domain experts

Each agent runs

Read the issue exit criteria and the diff.

Evaluate from its assigned perspective.

Report confidence (high / medium / low) per finding.

Return verdict: approve, comment, or request-changes.

Synthesis

Preserve dissent — surface contradictions verbatim. The human author decides.

Triage by confidence — investigate low-confidence findings before posting.

Deduplicate findings across agents.

Run tests: make check

Post a single review on the merge request, attributing each finding to its perspective.

Code-quality escalation

Severity

Action

Blocks correctness (bug, data loss)

request-changes

Introduced by this PR

request-changes

Pre-existing but touched

comment + new ticket

Pre-existing and untouched

investigate → ticket if warranted

Review PR $ARGUMENTS — multi-perspective agent review

Spin multiple agents in parallel, each with a distinct perspective. Run all agents in fresh contexts.

Setup

Read the issue linked to the PR. Note the exit criteria.
Read the diff of the merge request.
Assess risk level and determine proportional depth (see table below).
Launch review agents in parallel:

Agent	Focus	Key question
Correctness	Logic, edge cases, test coverage	Does this do what the exit criteria say?
Consistency	Style, naming, docs, stale references	Does this fit the rest of the codebase?
Scope	Over-engineering, unrelated changes	Does this change only what the ticket asks?
Red team	Adversarial inputs, broken invariants	How can this break?
Doc propagation	Downstream text accuracy	Do docs and configs still match the code?

Proportional depth

PR risk	Agents
Trivial + user present	Skip PR — merge directly
Trivial (typo, config)	Correctness only
Standard	Correctness + Consistency
Standard + scripts	+ Doc propagation
Substantial	All five
High-risk (schema, methodology)	All five + domain experts

Each agent runs

Read the issue exit criteria and the diff.
Evaluate from its assigned perspective.
Report confidence (high / medium / low) per finding.
Return verdict: approve, comment, or request-changes.

Synthesis

Preserve dissent — surface contradictions verbatim. The human author decides.
Triage by confidence — investigate low-confidence findings before posting.
Deduplicate findings across agents.
Run tests: make check
Post a single review on the merge request, attributing each finding to its perspective.

Code-quality escalation

Severity	Action
Blocks correctness (bug, data loss)	request-changes
Introduced by this PR	request-changes
Pre-existing but touched	comment + new ticket
Pre-existing and untouched	investigate → ticket if warranted