review

Argus Code Review

Multi-agent code review orchestrator with cross-validation from multiple LLM providers.

---

Configuration

Step 1: Parse Arguments and Load Config

1. Parse $ARGUMENTS for:
   • --level <fast|balanced|comprehensive> - Skip confirmation prompt
   • --scope <path> - Override scope directory
   • --output <format> - Override output format
   • --yes or -y - Accept defaults without confirmation
   • --external - Detect external validators and add any detected tools to validators
2. Load defaults.json for review levels and options
3. Default to balanced level

Step 2: Confirmation or Execute

If --level OR --yes provided: Skip to Step 3 (no prompt)

Otherwise, show confirmation prompt:

Argus Code Review
=================
Ready to run with:
  Level:  Balanced (Sonnet, ~5 min)
  Scope:  {detected_scope}
  Output: inline

[1] Change level  [2] Change scope  [3] Change output  [Enter] Start  [Q] Cancel

Handle input:

• Enter/empty → proceed with shown settings
• 1 → prompt for level (fast/balanced/comprehensive)
• 2 → prompt for scope path
• 3 → prompt for output format (inline/markdown/json)
• Q/q → abort review

Step 3: Apply Selected Level

Load the settings from the selected review level:

• model - Claude model for agents
• depth - Which phases to run
• strictness - Issue sensitivity threshold
• validators - Validation providers
• output - Output format (can be overridden by --output)

If --external is set, detected tools are appended to validators in Step 6.

Step 4: Smart Scope Detection

If scope is "auto" (default), detect in order:

1. Check if src/ exists → use src/
2. Check if lib/ exists → use lib/
3. Check if app/ exists → use app/
4. Check if packages/ exists → use packages/ (monorepo)
5. Fall back to . (current directory)

Step 5: Display Summary and Start

Argus Code Review
=================
Level:  {selected level}
Model:  {config.model}
Scope:  {detected or config.scope}
Phases: {phases based on depth}

Starting review...