qlint

Multi-language code quality scanner. Walks any codebase and produces JSON, HTML, and Markdown reports covering complexity, duplication, security, code smells, and predictive risk (which files are most likely to cause your next outage).

Install

# from PyPI
pip install qlint

# from source
git clone https://github.com/ropean/qlint.git
cd qlint
pip install -e .

Usage

qlint /path/to/repo                  # JSON + HTML + Markdown; auto-opens HTML
qlint /path/to/repo --format md      # Markdown only
qlint /path/to/repo --format json,md # JSON + Markdown, no HTML, no auto-open
qlint /path/to/repo --no-open
qlint /path/to/repo --json-only      # print JSON to stdout, no files
qlint /path/to/repo -v               # verbose per-file output
qlint                                # interactive: prompts for path

qlint /path/to/repo --risk           # predictive risk report (top 5 chunked narrative)
qlint /path/to/repo --risk --risk-window 30
qlint /path/to/repo --risk-md /tmp/risk.md   # standalone markdown risk report

Reports are written to ~/Downloads/qlint-reports/<repo>/ as report.json, report.html, and report.md. HTML auto-opens only when an HTML report is produced. --risk alone prints to the console and writes nothing.

Options

Flag	Description
--format, -f	Comma-separated formats: json, html, md (default: all three)
--output, -o	Custom JSON output path (implies json)
--html	Custom HTML output path (implies html)
--md	Custom Markdown output path (implies md)
--risk	Print predictive risk report (top 5 files most likely to break)
--risk-window	Days of git history to consider as "recent" (default 90)
--risk-md	Write standalone risk Markdown report to PATH
--json-only	Skip files, print JSON to stdout
--no-open	Do not auto-open HTML report
--verbose, -v	Show per-file progress
--version	Show version

Analysis Features

Feature	What it detects
Complexity	Cyclomatic complexity per function, flags > 10
Duplication	Duplicate 6-line blocks across files
Security	Hardcoded secrets, dangerous functions (eval, exec, …)
Code Smells	Long functions, deep nesting, long parameter lists
Predictive Risk	Recency-weighted: recent_churn × complexity × (1 + 2·bug_fix_ratio) × (1 + 0.15·authors) / 100. Each top file gets a risk_level (critical/high/medium/low), structured signals, and human-readable reasons.

Predictive Risk

qlint --risk highlights the files most likely to cause your next outage by combining:

• Static signals — complexity, smells, security issues
• Git signals (recent window) — churn, commit frequency, bug-fix ratio, author count

Bug-fix detection scans commit subjects for fix|bug|hotfix|patch|revert (case-insensitive). Files with a high share of recent bug-fix commits get amplified scores — past fix density is a strong predictor of future fixes.

Risk levels are bucketed: critical (≥50), high (≥20), medium (≥5), low (<5).

Languages

Python, JavaScript, TypeScript, Java, Go, Ruby, Rust, C, C++, C#, PHP, Swift, Kotlin, Scala, Shell, HTML, CSS, SQL, YAML, JSON, and more.

JSON Output

{
  "scanId": "...",
  "timestamp": "...",
  "repository": { "path": "...", "totalFiles": 42, "totalLines": 8500, "languages": {} },
  "files": [
    {
      "path": "src/main.py",
      "language": "Python",
      "metrics": { "loc": 120, "code": 95, "comments": 10, "blank": 15, "functions": 8, "classes": 2, "complexity": 4.2 },
      "smells": [],
      "security_issues": [],
      "git_risk": {
        "commits": 12, "churn": 340, "authors": 2,
        "risk_score": 25.5, "risk_level": "high",
        "signals": { "...": "see gitRisk.top_risk_files for the full shape" },
        "reasons": ["active churn (340 lines / window)"]
      }
    }
  ],
  "qualityScore": 87,
  "grade": "B",
  "gitRisk": {
    "available": true,
    "window_days": 90,
    "top_risk_files": [
      {
        "file": "src/main.py",
        "risk_score": 25.5,
        "risk_level": "high",
        "reasons": ["active churn (340 lines / window)"],
        "signals": {
          "complexity": 4.2,
          "recent_commits": 5,
          "recent_churn": 340,
          "all_time_commits": 12,
          "all_time_churn": 600,
          "authors": 2,
          "bug_fix_ratio": 0.4,
          "smells": 1,
          "security_issues": 0
        }
      }
    ]
  }
}

Claude Code skill — qlint-refactor

The .claude/skills/qlint-refactor/ directory ships a Claude Code skill that turns the scanner into a feedback signal: it picks the riskiest file, proposes a behavior-preserving refactor, re-scans, and verifies the score dropped.

Loop:

1. Baseline via scripts/risk_top.py (top-N risk JSON)
2. Locate the worst function in the top-1 file
3. Propose a minimal refactor (extract function, flatten conditionals, lookup tables)
4. Apply, then scripts/verify.py FILE --baseline N.NN → improved / unchanged / regressed
5. Roll back on no-improve; max 3 attempts per file; exit when all top-5 are risk_score < 50

Activates on prompts like "list top risk files", "explain why this file is risky", "refactor the risky code", "reduce risk score", "run the risk loop".

Development

make install   # pip install -e ".[dev]"
make lint      # ruff check
make format    # ruff format
make build     # python3 -m build

Requirements

• Python 3.11+
• git in PATH (for predictive risk; optional — --risk degrades gracefully on non-git dirs)