opencode-review-helper

OpenCode plugin for reviewing AI-generated code changes. Helps humans efficiently review large changesets by suggesting optimal review order and identifying potential breaking changes.

Features

- Review Order: Analyzes changed files and suggests optimal review sequence based on dependencies, file type priority, and complexity
- Impact Analysis: Finds code outside the changeset that could be affected by the changes (direct consumers, transitive dependencies, test coverage gaps)

Installation

``bash
bunx opencode-review-helper setup
`

This will:
1. Add the plugin to your ~/.config/opencode/opencode.jsonc
2. Let you choose which model to use for the impact-explorer sub-agent (default: google/gemini-3-flash)

Restart opencode after setup to load the plugin. The plugin uses @latest so it will auto-update when you restart OpenCode.

Usage

Direct Slash Command Use

Run the slash command:
`
/review-order
`

This runs the full workflow:
1. Analyzes changed files and determines initial review order
2. Finds external files impacted by the changes
3. Merges everything into a single prioritized list

Output is a unified table showing all files to review, with changed files first (prioritized by dependencies and impact) followed by external files that may need attention.

Agent Use

Tab twice to go to Review-Helper mode:


Enter your prompt:
`
Take a look at everything since the 0.2.0 release as if it were a single PR. Guide me through how to review it.
`

Output:
`
Review Order for v0.2.0 → HEAD
| # | File | Reason | Score |
|---|------|--------|-------|
| 1 | src/tools/review-order.ts | Core scoring/sorting logic; foundation for all agents | 80 |
| 2 | src/agents/review-order.ts | Sub-agent wrapping review_order tool | 70 |
| 3 | src/agents/impact-explorer.ts | Sub-agent wrapping impact_analysis tool | 70 |
| 4 | src/agents/review-helper.ts | Orchestrator coordinating sub-agents | 60 |
| 5 | src/agents/index.ts | Barrel export for agent definitions | 50 |
| 6 | src/index.ts | Plugin entry point; command registration | 40 |
| 7 | package.json | Version bump 0.2.0 → 0.3.0 | 30 |
| 8 | AGENTS.md | Local testing/release docs | 20 |
| 9 | README.md | User-facing docs reorganization | 10 |
| 10 | agents/*.md (deleted) | Removed markdown agent files | 5 |
---
Summary of Changes
Theme: Subagent architecture for /review-order slash command
Structural Changes
- Added: src/agents/ directory with 4 new files
- Deleted: agents/*.md files (agent prompts moved to TypeScript)
- Refactored: src/index.ts significantly simplified (136 lines → ~30 lines of logic)
Key Review Points
| File | What to Check |
|------|---------------|
| src/tools/review-order.ts | +139 lines. New scoring algorithm? Dependency detection? |
| src/agents/review-helper.ts | Orchestration flow: review-order → impact → merge |
| src/index.ts | Verify plugin registration still works after major trim |
| agents/*.md deletion | Confirm prompts weren't lost, just moved to .ts |
Testing Gaps
⚠️ No test files for any changed code. Manual testing via /review-order command required.
---
Want me to read specific files for deeper analysis?
`

Agents

Three agents available for direct use or extension:

| Agent | Purpose |
|-------|---------|
| review-helper | Orchestrator - runs full /review-order workflow |
| review-helper:review-order | Determines optimal file review order |
| review-helper:impact-explorer | Finds external impacted code |

The orchestrator does NOT make up its own recommendations—it only presents what the subagents return. It also does NOT automatically run tests or apply fixes.

$3

Run impact analysis only:
`bash
opencode run --agent review-helper:impact-explorer "Analyze impact of current changes"
`

Get review order without impact merge:
`bash
opencode run --agent review-helper:review-order "Get review order for changed files"
`

Configuration

Create ~/.config/opencode/review-helper.json or .opencode/review-helper.json in your project:

`json
{
"models": {
"explorer": "google/gemini-3-flash"
},
"review_order": {
"type_priority": [
"migration",
"schema",
"model",
"service",
"..."
],
"instructions": "Always review database changes before business logic"
},
"impact_analysis": {
"max_depth": 2,
"max_results_per_level": 50,
"exclude_patterns": ["/.test.ts", "/__tests__/*"]
}
}
`

Default type_priority includes 18 keywords. See src/config.ts for full list.

$3

| Option | Description | Default |
|--------|-------------|---------|
| models.explorer | Model for sub-agent exploration | google/gemini-3-flash |
| review_order.type_priority | File type keywords in priority order | See config.ts |
| review_order.instructions | Custom ordering instructions | - |
| impact_analysis.max_depth | Transitive analysis depth (1-3) | 2 |
| impact_analysis.max_results_per_level | Max results per category | 50 |
| impact_analysis.exclude_patterns` | Glob patterns to skip | Test files |

License

MIT