Reassigning Your AI Coder to Security Detail
Agent skills that turn your existing LLM coding assistant into a parallelized SAST scanner without extra tooling.

What it does
sast-skills is a set of markdown orchestration files—CLAUDE.md for Claude Code, AGENTS.md for the rest—that repurpose your AI assistant’s agent mode to audit code for security flaws. It maps the target’s architecture and data flows, then dispatches thirteen parallel subagent tasks to hunt for vulnerability classes like SQL injection, SSRF, and business logic bugs. Finally, it synthesizes a single severity-ranked report with proof-of-concept details and remediation guidance, all using the model you already pay for.
The interesting bit
Instead of wrapping a traditional scanner, the project treats the LLM itself as the analysis engine. Each vulnerability class follows a two-phase reconnaissance-then-verification approach, and the orchestration file manages the entire workflow automatically, skipping any phase whose output already exists so you can iterate without redundant work.
Key highlights
- Covers 13 vulnerability classes, from SQL injection and GraphQL injection to insecure file uploads and JWT implementation flaws.
- Runs fully inside Claude Code, Cursor, Opencode, or Codex with no third-party dependencies.
- Parallelizes detection across subagents before consolidating findings into a ranked final report.
- Automatically resumes partially completed scans by checking for existing output files in the
sast/folder. - Recommends Claude Code with the Opus model, but explicitly supports other IDEs and models for cost-conscious users.
Caveats
- Any existing
CLAUDE.mdorAGENTS.mdin your project will conflict with the toolkit’s orchestration file. - The workflow expects the target codebase to be opened inside a dedicated
sast-filesworkspace, which may be awkward for projects that already rely on their own agent markdown.
Verdict
Developers who already live inside an AI-assisted IDE and want a quick, zero-install security sniff-test should give this a spin. If you need deterministic, compliance-grade scanning with rigorously quantified false-positive rates, you are still better served by conventional SAST tooling.
Frequently asked
- What is utkusen/sast-skills?
- Agent skills that turn your existing LLM coding assistant into a parallelized SAST scanner without extra tooling.
- Is sast-skills open source?
- Yes — utkusen/sast-skills is open source, released under the MIT license.
- How popular is sast-skills?
- utkusen/sast-skills has 1.2k stars on GitHub and is currently accelerating.
- Where can I find sast-skills?
- utkusen/sast-skills is on GitHub at https://github.com/utkusen/sast-skills.