Your AI Becomes a Security Team. Every Language. Every Layer. Zero Tools.
npx skills add ersinkoc/security-checkThat's it. Open your AI assistant and say "run security check".
WrongStack is a free, open-source AI coding agent with a Brain, a Memory, and a full toolbox. It reads your code, edits files, runs commands, and coordinates specialist agents — across six surfaces, from a plain terminal REPL to a cross-machine HQ dashboard. No subscription required, and you keep your hand on every permission.
curl -fsSL https://wrongstack.com/install.sh | sh # macOS / Linux — self-contained binaryirm https://wrongstack.com/install.ps1 | iex # Windows (PowerShell) — no Node.js needed| What you get | |
|---|---|
| 🌐 200+ LLM providers | Catalog pulled live from models.dev — Anthropic, OpenAI, Google, plus OAuth sign-in for Claude Pro/Max, ChatGPT and Copilot, and any OpenAI-compatible endpoint (Ollama, vLLM, LM Studio) |
| 🛠️ 67 built-in tools | Edits, lint/typecheck/test, execution, git, web, browser/E2E and a SQLite codebase index — every call gated by per-tool permissions |
| 🧠 SAGE memory | Project-wide long-term memory in SQLite/FTS5, anchored to files, symbols and commits — it gets better at your codebase over time |
| 🖥️ Six surfaces | Readline REPL · Ink/React TUI (--tui) · WebUI · SimpleUI · Desktop · cross-machine HQ (--hq) |
| 🤖 Fleet orchestration | A Director fans out specialist subagents over a project mailbox; eternal & parallel goal loops run until the contract verifies |
| 🔍 Chimera & Kanban | Auto-review agents that critique your diffs with severity-ranked findings, plus durable Kanban boards with atomic verification |
| 🔐 Secure by default | Encrypted secrets at rest, project-root containment, opt-in YOLO mode — MIT licensed, TypeScript-strict |
📊 The perfect pairing: This band tells you exactly where your Claude limits stand — and WrongStack keeps you moving when they close in. It reads plan windows for Claude, ChatGPT, Copilot, Z.AI and more right in its own statusline and quota page, and when one provider runs dry, fallback chains rotate you onto the next model automatically. Watch the band, dodge the wall.
Shell script (macOS/Linux)
curl -fsSL https://raw.githubusercontent.com/ersinkoc/security-check/main/skills.sh | bashWith options:
# Install only specific categories
curl -fsSL https://raw.githubusercontent.com/ersinkoc/security-check/main/skills.sh | bash -s -- --category injection server
# Install only specific languages
curl -fsSL https://raw.githubusercontent.com/ersinkoc/security-check/main/skills.sh | bash -s -- --lang go typescript python
# List all available categories
curl -fsSL https://raw.githubusercontent.com/ersinkoc/security-check/main/skills.sh | bash -s -- --listPowerShell (Windows)
irm https://raw.githubusercontent.com/ersinkoc/security-check/main/skills.ps1 | iex Manual installation
git clone https://github.com/ersinkoc/security-check.git
cp security-check/scan-target/CLAUDE.md /path/to/your/project/
mkdir -p /path/to/your/project/.claude/skills
mkdir -p /path/to/your/project/.claude/skills/security-check
cp security-check/SKILL.md /path/to/your/project/.claude/skills/security-check/SKILL.md
cp -r security-check/skills/sc-* /path/to/your/project/.claude/skills/A collection of 51 agent skills following the agentskills.io standard that transforms your AI coding assistant into a comprehensive security scanning team. No binaries, no dependencies, no CI pipeline changes — just natural language prompts.
It is not a traditional SAST tool. It uses LLM reasoning to understand code context, trace data flows across function boundaries, and evaluate framework-level protections. The result: fewer false positives and deeper analysis than pattern-matching tools.
| Platform | Install via skills.sh | Manual |
|---|---|---|
| Claude Code | npx skills add ersinkoc/security-check |
Copy .claude/skills/ + CLAUDE.md |
| Cursor | npx skills add ersinkoc/security-check |
Copy .agents/skills/ + AGENTS.md |
| Codex | npx skills add ersinkoc/security-check |
Copy .agents/skills/ + AGENTS.md |
| Gemini CLI | npx skills add ersinkoc/security-check |
Copy .agents/skills/ + AGENTS.md |
| OpenCode | npx skills add ersinkoc/security-check |
Copy .agents/skills/ + AGENTS.md |
| Windsurf | npx skills add ersinkoc/security-check |
Copy .agents/skills/ + AGENTS.md |
| Roo Code | npx skills add ersinkoc/security-check |
Copy .agents/skills/ + AGENTS.md |
| Amp | npx skills add ersinkoc/security-check |
Copy .agents/skills/ + AGENTS.md |
See docs/SUPPORTED_PLATFORMS.md for detailed platform-specific setup.
After installation, tell your AI assistant:
| Command | What it does |
|---|---|
| "run security check" | Full 4-phase security audit |
| "scan diff" or "PR scan" | Scan only changed files |
| "scan for vulnerabilities" | Full scan (alias) |
| "check changes for security" | Diff mode (alias) |
Open Claude Code's / menu and select /security-check. The launcher asks for
the unresolved audit profile, scope, validation mode, and prior-report behavior using
native selection cards. Explicit choices in the original prompt are not asked again.
The npx skills installation exposes the repository as the security-check bundle.
The shell and manual installers additionally expand the 51 scanning skills into sibling
directories while installing the same /security-check launcher.
Default choices are non-destructive: Standard, Whole repository, Source only, and Continue and revalidate. Selecting Changed files switches to diff mode; selecting Replace explicitly authorizes replacement of the previous report.
Each skill follows the Agent Skills specification:
sc-sqli/
├── SKILL.md # Metadata + detection instructions
└── references/ # Checklists (for language skills)
| Category | Skills | What it detects |
|---|---|---|
| Core Pipeline | sc-orchestrator, sc-recon, sc-dependency-audit, sc-verifier, sc-report, sc-diff-report | Pipeline orchestration, architecture mapping, supply chain, verification, reporting |
| Injection (9) | sc-sqli, sc-nosqli, sc-graphql, sc-xss, sc-ssti, sc-xxe, sc-ldap, sc-cmdi, sc-header-injection | SQL/NoSQL/GraphQL injection, XSS, SSTI, XXE, LDAP, command injection, CRLF |
| Code Execution (2) | sc-rce, sc-deserialization | eval/exec RCE, pickle/ObjectInputStream/unserialize/BinaryFormatter |
| Access Control (4) | sc-auth, sc-authz, sc-privilege-escalation, sc-session | Broken auth, IDOR, privilege escalation, session fixation |
| Data Exposure (3) | sc-secrets, sc-data-exposure, sc-crypto | Hardcoded secrets, PII leaks, weak cryptography |
| Server-Side (4) | sc-ssrf, sc-path-traversal, sc-file-upload, sc-open-redirect | SSRF, directory traversal, file upload, open redirect |
| Client-Side (4) | sc-csrf, sc-cors, sc-clickjacking, sc-websocket | CSRF, CORS misconfig, clickjacking, WebSocket hijacking |
| Logic (3) | sc-business-logic, sc-race-condition, sc-mass-assignment | Business logic flaws, TOCTOU, mass assignment |
| API (3) | sc-api-security, sc-rate-limiting, sc-jwt | OWASP API Top 10, rate limiting, JWT flaws |
| Infrastructure (3) | sc-iac, sc-docker, sc-ci-cd | IaC misconfig, Docker security, CI/CD pipeline vulns |
| Languages (7) | sc-lang-go, sc-lang-typescript, sc-lang-python, sc-lang-php, sc-lang-rust, sc-lang-java, sc-lang-csharp | Language-specific deep scanning with 400+ item checklists |
| Specialized Surfaces (3) | sc-ai-security, sc-protocol-security, sc-local-ipc | LLM/agent tools and memory, RPC/brokers/webhooks, desktop/mobile/local IPC |
Each language scanner uses a dedicated 400+ item checklist with specific CWE references:
| Language | Skill | Key Focus Areas |
|---|---|---|
| Go | sc-lang-go | unsafe package, goroutine leaks, race conditions, crypto/rand, template XSS |
| TypeScript/JS | sc-lang-typescript | Prototype pollution, eval injection, DOM XSS, npm supply chain, Next.js |
| Python | sc-lang-python | pickle RCE, SSTI, subprocess injection, Django/Flask/FastAPI |
| PHP | sc-lang-php | unserialize gadgets, phar, type juggling, Laravel/WordPress |
| Rust | sc-lang-rust | unsafe blocks, FFI, integer overflow, Send/Sync, serde bombs |
| Java/Kotlin | sc-lang-java | Deserialization, JNDI injection, Spring SpEL, HQL, XXE |
| C#/.NET | sc-lang-csharp | BinaryFormatter RCE, EF raw SQL, Blazor JS interop, SignalR |
PHASE 1: RECON Architecture mapping, tech stack detection, dependency audit
│
▼
PHASE 2: HUNT 40+ vulnerability skills run in parallel
│
▼
PHASE 3: VERIFY False positive elimination, confidence scoring (0-100)
│
▼
PHASE 4: REPORT CVSS severity classification, remediation roadmap
│
▼
OUTPUT: security-report/SECURITY-REPORT.md
The four phases are evidence-led: reconnaissance seeds a coverage ledger, hunters
produce candidates, and a separate verification pass tries to disprove each candidate.
Only a demonstrated trust-boundary failure is confirmed; unresolved source-grounded
leads remain needs_validation without severity. See
Evidence and Coverage Model.
After a scan, the security-report/ directory contains:
| File | Description |
|---|---|
SECURITY-REPORT.md |
Final consolidated security assessment |
architecture.md |
Codebase architecture map |
dependency-audit.md |
Supply chain analysis |
verified-findings.md |
Findings after verification |
coverage-ledger.md |
Explicit covered, candidate, blocked, deferred, and excluded units |
findings.json |
Structured final verdicts: confirmed, needs_validation, rejected |
findings/*.json |
Raw findings from each skill |
- Copy
templates/SKILL_TEMPLATE.mdfor vulnerability skills - Copy
templates/LANG_SKILL_TEMPLATE.mdfor language skills - Follow the agentskills.io specification
See docs/SKILL_DEVELOPMENT_GUIDE.md for detailed instructions.
See CONTRIBUTING.md for guidelines.
MIT License — see LICENSE for details.
The evidence and coverage model in v1.2 was informed by Cloudflare's MIT-licensed
security-audit-skill and
adapted to security-check's multi-skill four-phase pipeline.
Ersin Koc — ECOSTACK TECHNOLOGY OU