AIDE - AI Coding Quality Inspector

Detect hallucinations, stub code, security issues, and errors in AI-generated code — zero config, zero LLM cost by default.

Features

• 14 stable detectors (16 total with 2 preview) covering AI-generated code pitfalls: package hallucination, API hallucination, stub functions, empty implementations, hardcoded values, fake URLs, swallowed errors, unhandled promises, unused declarations, unreachable code, resource leaks, security, structure, weak validation, and narrative comments
• 13 languages: Python, TypeScript, JavaScript, Go, Java, Rust, Ruby, PHP, C, C++, C#, Swift, Kotlin
• Pure local scan — aide scan runs entirely offline, 0 token consumption
• LLM-assisted review — aide scan-llm uses an independent LLM to reduce false positives (40-60% pre-filtered locally before LLM calls)
• Configurable — .aiderc.json project config, CLI flags override, strict mode, confidence filtering
• CI-friendly — --exit-code flag, JSON/supervisor output formats, --offline mode

Install

npm install -g @aide-dev/aide

Requires Node.js >= 20.

Quick Start

# Scan your project (pure local, no LLM needed)
aide scan

# Strict mode: only high + medium severity
aide scan --strict

# LLM-assisted scan (reduces false positives)
aide scan-llm --model gpt-4o

# Configure LLM provider
aide configure-llm

# JSON output for CI integration
aide scan --format json --exit-code

Commands

Command	Description
aide scan	Pure local scan, 0 token cost
aide scan --auto	Auto-enable LLM reduce-fp if model configured, filter low severity
aide scan --strict	Only high + medium severity, filter low confidence
aide scan-llm	LLM-assisted scan for lower false positive rate
aide configure-llm	Configure LLM provider (OpenAI, Anthropic, etc.)

Common Options

--strict              Only report high + medium severity (filters low confidence)
--auto                Auto-enable LLM reduce-fp if model configured, exclude low severity
--preview             Enable preview detectors (narrative-comments, dead-code)
--skip <rules>        Disable specific detectors
--only <rules>        Only run specific detectors
--format <fmt>        Output: default, verbose, ai, json, supervisor
--offline             Disable network requests
--exit-code           Exit with code 1 if issues found (CI mode)
--min-confidence <lv> Minimum confidence: high, medium, low

Detectors

Detector	What it catches
package-hallucination	Imports of non-existent npm/PyPI/Go/Java packages
api-hallucination	Calls to non-existent stdlib/module APIs
stub-function	Functions returning only true/false/null/0/{}/[]
empty-impl	Empty function/class method bodies
hardcoded-value	Hardcoded URLs, IPs, ports, paths
fake-url	Placeholder/example URLs in production code
swallowed-error	catch/except blocks with no error handling
unhandled-promise	Floating promises, missing .catch()
unused-declaration	Unused variables and imports
unreachable-code	Code after return/throw/break
resource-leak	Unclosed files, connections, streams
security	Hardcoded secrets, SQL injection, XSS, command injection
weak-validation	Weak input validation patterns
structure	Project structure and organization issues
narrative-comments ⚡	Narrative/inline-comment code that should be real logic (preview)
dead-code ⚡	Dead/unreachable code detection (preview)

⚡ = preview detectors, enable with --preview flag

Configuration

Create .aiderc.json in your project root:

{
  "skipRules": ["narrative-comments"],
  "strict": true,
  "ignore": ["vendor"],
  "format": "supervisor",
  "minConfidence": "medium"
}

Priority: CLI flags > .aiderc.json > defaults.

How It Works

Source files → Local detectors (regex + AST + registry)
                → isInNonCode filter (skip strings/comments/regex)
                → Local prefilter (13 rules, 0 token)
                → [Optional] LLM review (rule-customized context)
                → Severity + confidence filtering
                → --strict mode removes low confidence + low severity
                → --auto mode auto-enables reduce-fp if LLM configured
                → Results

Development

git clone https://github.com/ldkong2077/aide.git
cd aide
npm install
npm run build
npm test

License

MIT