review-code

name	review-code
description	Use when reviewing existing code against development skill best practices — orchestrates multi-agent domain reviews for thoroughness
user-invocable	true

Review Code

Review code against all development skill best practices — with sub-agent orchestration for thoroughness.

When to Use

Use this skill when:

• Reviewing an existing project for compliance with skill-defined patterns
• Checking a specific package or domain against best practices
• Running a full multi-agent code review

Review Domains

Reviews are organized into 5 domains, each with its own check tables:

Domain	Reference File	Categories	Skills
Go Foundations	./references/review-domain-go-foundations.md	1-4 (Layout, Principles, Modern Go, Logging, Utils)	go-foundations
Go CLI	./references/review-domain-go-cli.md	5-7 (Cobra, Commands, Output Lifecycle, TUI)	go-cli
Go Backend & Frontend	./references/review-domain-go-backend-frontend.md	8-9 (Architecture, Assets)	go-backend, go-frontend
Go Concurrency	./references/review-domain-go-concurrency.md	10-11 (Concurrency, Pipeline)	go-concurrency
Infrastructure	./references/review-domain-infra.md	12-15 (CI/CD, Chrome, README, Deps)	project-ci-cd, chrome-extension-basics, project-readme, go-foundations

Reference paths in this skill are relative to this SKILL.md. Sibling skills (e.g. go-cli, go-foundations) live alongside this skill, so their files are at ../<skill-name>/SKILL.md and ../<skill-name>/references/<file>.md.

---

Workflow

Step 1: Detect Project Type

Inspect the current directory to determine what kind of project this is:

Indicator	Project Type
go.mod + cmd/, utils/, no internal/server/	Go CLI Only
go.mod + internal/server/static/, no utils/	Go Web Only
go.mod + internal/server/static/ and utils/	Go CLI + Web (hybrid)
manifest.json with manifest_version field	Chrome Extension
go.mod only (minimal)	Go Project (treat as CLI Only)

If no recognizable project structure is found, report that and stop.

Step 2: Parse Target Argument

If no target argument was provided, go to Step 3a (Full Review).

If a target argument was provided:

2a. Check keyword table:

Keyword(s)	Domain	Subset
foundations, core, layout, logging, utils	Go Foundations	All categories
cli, cobra, commands, tui	Go CLI	All categories
backend, frontend, server, web	Go Backend & Frontend	All categories
concurrency, pipeline, highway, goroutines	Go Concurrency	All categories
cicd, ci, cd, docker, makefile	Infrastructure	CI/CD category only
chrome, extension	Infrastructure	Chrome category only
readme	Infrastructure	README category only
deps, dependencies	Infrastructure	Deps category only
infra, infrastructure	Infrastructure	All categories

If a keyword matches, go to Step 3b (Targeted Review) with the mapped domain and subset.

2b. Check if package path:

If the target contains /, treat it as a package path. Map to domain(s):

Path Pattern	Domain(s)
cmd/**	Go CLI
internal/server/static/**	Go Backend & Frontend
internal/server/**	Go Backend & Frontend
internal/highway/** or internal/display/**	Go Concurrency
utils/**	Go Foundations
internal/** (anything else)	Go Foundations + Go Backend & Frontend
.github/**, Makefile, Dockerfile, README.md	Infrastructure

Go to Step 3b (Targeted Review) with the mapped domain(s).

2c. Validate applicability:

If the target maps to a domain that does not apply to the detected project type (e.g., backend for a Go CLI Only project), report:

The [domain] domain does not apply to this project type ([project type]).

And stop.

---

Step 3a: Full Review (Multi-Agent Orchestration)

Determine which domains apply based on project type:

Project Type	Applicable Domains
Go CLI Only	Go Foundations, Go CLI, Go Concurrency (conditional), Infrastructure
Go Web Only	Go Foundations, Go CLI, Go Backend & Frontend, Infrastructure
Go CLI + Web (hybrid)	Go Foundations, Go CLI, Go Backend & Frontend, Go Concurrency (conditional), Infrastructure
Chrome Extension	Infrastructure

If only one domain applies (e.g., Chrome Extension), skip sub-agents and handle it directly — read the domain reference file, load its skills, run checks, and generate the report inline. Go to Step 4.

If multiple domains apply, launch parallel sub-agents. For each applicable domain, launch a Task with these parameters:

• subagent_type: "general-purpose"
• model: "sonnet"

Before building the prompt, resolve [SKILLS_DIR] to an absolute path: it is the directory that contains the sibling skills — i.e. the parent of this review-code skill directory (the directory holding this SKILL.md, then one level up). Sub-agents run in the target project's working directory, so they need absolute paths to locate skill files.

Use this prompt template for each sub-agent. Substitute [SKILLS_DIR] with its resolved absolute value, and fill in all other bracketed values:

You are a focused code review agent for the [DOMAIN_NAME] domain.

## Context
- Project type: [PROJECT_TYPE]
- Working directory: [CWD]
- Skills directory (absolute path, for locating skill/reference files): [SKILLS_DIR]

## Instructions

1. Read your domain check tables:
   Read the file at [SKILLS_DIR]/review-code/references/review-domain-[DOMAIN_FILE].md

2. Read the skill files listed in your domain reference header. The skill files are at:
   [SKILLS_DIR]/[SKILL_NAME]/SKILL.md
   Also read any reference files mentioned within the skills (reference files within a skill are at [SKILLS_DIR]/[SKILL_NAME]/references/).

3. Thoroughly inspect the project in [CWD] against EVERY check in your domain reference:
   - Use Read, Glob, Grep, and Bash (read-only) to verify each check
   - Only check categories marked as applicable to [PROJECT_TYPE]
   - For conditional categories, first detect if the pattern exists before deep-diving
   - Read the ACTUAL source files — do not guess or assume

4. Report findings using the output format defined in your domain reference file.

5. HARD BOUNDARY: Do not flag anything not defined in your check tables or loaded skill files.
   If you cannot cite a specific skill section, it is NOT a finding.

6. End your response with exactly:
   SUMMARY_LINE: categories_checked=N pass=N issues=N skipped=N total_issues=N

Launch ALL domain sub-agents in a single message (parallel tool calls) for maximum throughput.

Go to Step 4.

---

Step 3b: Targeted Review (Single Agent, Inline)

For targeted reviews, do NOT launch sub-agents. Handle the review directly:

1. Read the domain reference file: ./references/review-domain-[domain].md
2. Load the skill files listed in the reference header (sibling skills at ../[SKILL_NAME]/SKILL.md)
3. If a subset was specified (e.g., keyword cicd maps to only the CI/CD category within Infrastructure), check only that category — skip the rest in the reference file
4. If a package path was specified, narrow checks to patterns relevant within that package scope (skip file-existence checks outside the package, but still check cross-cutting concerns like error handling and logging patterns)
5. Run all applicable checks and generate the report

Go to Step 4.

---

Step 4: Generate Review Report

For full reviews with sub-agents: After all Task calls complete, collect their outputs and combine into a unified report.

For targeted reviews: You already have the findings from Step 3b.

Present the final report in this format:

## Code Review Report: [Project Name]

**Project type:** [CLI Only | Web Only | CLI + Web | Chrome Extension]
**Review scope:** [Full | Targeted: domain-name | Targeted: package-path]
**Skills checked against:** [list of loaded skills]

---

[Domain sections — each sub-agent's output, or inline findings for targeted reviews]

---

### Summary

| Category | Status | Issues |
|----------|--------|--------|
| Project Layout | PASS | 0 |
| Core Principles | ISSUES | 2 |
| Logging | PASS | 0 |
| ... | ... | ... |

**Total issues found:** N

Step 5: Offer to Fix

After presenting the report, ask:

Would you like me to fix these issues? I can address them one category at a time.

If the user agrees, work through each category's issues in order, making the changes described in the "Fix" suggestions.

---

Key Principles

• Skills are the single source of truth — Every finding MUST cite the specific skill and section it comes from. If no skill defines a pattern, it is NOT a finding.
• Only check relevant categories — Skip categories that don't apply to the detected project type.
• Be specific — "Missing PrintFatal function in printer.go" not "Utils package is incomplete."
• Reference fixes to skill patterns — Don't invent best practices. Every "Expected" and "Fix" must come from a loaded skill.
• Don't flag what skills don't cover — No findings for: missing linting, missing formatting, missing pre-commit hooks, missing documentation beyond README.
• Detect before deep-diving — Check if a pattern is even used before reviewing it in depth (e.g., don't review concurrency if no goroutines exist).
• Sub-agents read actual code — Each domain agent must Read the actual source files, not guess based on file names.

Out of Scope (Hard Boundary)

The following are NOT defined in any loaded skill and MUST NOT be flagged as issues:

Category	Specific Examples
Linting & Formatting	No golangci-lint, no gofmt, inconsistent formatting
Pre-commit	No pre-commit hooks, no husky
Code Quality CI	No lint/format CI steps
Documentation beyond README	No godoc, no changelogs, no contributing guide
Docker Compose	No docker-compose for development
Database	No migrations, no schema files
Dependency tooling	No dependabot, no renovate
Security scanning	No SAST, no container scanning
Code style opinions	Naming conventions not in skills, personal preferences

Rule of thumb: If you are about to flag something and cannot point to a specific section in a loaded skill that defines the expected pattern, do not flag it.