Snyk's CLI scanner that discovers installed agents, MCP servers and skills on your machine and checks them for prompt injection and other risks.
A credible, signed security scanner from a known vendor, but it needs a Snyk token, uploads component content for analysis and can launch MCP servers during scans.
Agent Scan is a Python command-line tool, not a skill pack. It finds agent components on a machine (the README includes a support table of coding agents for macOS, Linux and Windows) and scans the MCP servers and skills it finds for threats such as prompt injection, sensitive-data handling and malware payloads hidden in natural language. Main commands are snyk-agent-scan scan, which analyzes components, and inspect, which lists them without analysis. A --no-skills flag skips skill analysis.
It is distributed through uvx or as a standalone binary from GitHub Releases, with checksums, an SBOM and GPG-signed release hashes; there is no npm package. It needs a Snyk account and an API token set in SNYK_TOKEN. To analyze components it sends information such as MCP server configurations, tool names and descriptions and skill content to the Agent Scan API, with secrets redacted first. To read tool descriptions it can start stdio MCP servers and contact remote ones, so interactive runs ask for consent per server and the README recommends running untrusted configs in a sandbox. The README warns that output formats, scores and field names are experimental and may change.
The skill count in the dossier comes from test fixtures under tests/, including intentionally malicious and invalid skills used to exercise the scanner, plus copies of public example skills. They are not meant to be installed.
A maintained scanner with CI and a large test suite, but it is a tool rather than a skill pack and its bundled skills are test fixtures.
Counted from the repository file tree. How we score depth.
Snyk's Agent Scan is a security scanner for the agents, skills and MCP servers configured on your machine; the repo we reviewed holds the README, docs and test fixtures, not the scanner code. Running it reads your local agent configuration and sends redacted component data, including skill content, to Snyk's API with your Snyk token, and it can contact your MCP servers (asking first in interactive runs, with an option to skip consent). The tool itself runs from uvx or a binary that was not available to review.
uvx snyk-agent-scan@<version> or a standalone binary; its source is not in the files we hold.README.md:83uvx snyk-agent-scan@latest ... (the README also shows a pinned @0.5.17 form).README.md:112--control-server can upload results elsewhere.README.md:284The repository (434 files) on Oct 29, 2026. Not a full review: 228 of 434 files were not downloaded (large repository; we reviewed the skill folders); it contains compiled binaries or archives we cannot read; we read 0 of 1 code files; it relies on code outside the repository: snyk-agent-scan package / binary (the scanner itself) and the Snyk Agent Scan API; The repo copy we hold contains only the README, docs and test fixtures (about 150 files, including sample skills and MCP servers used as scanner test cases); the scanner source is absent, and the pipeline's dossier and scan files for this slug were not available to cross-check.
A review, not a certification. Skills change after we check them — read the files before installing.
Apache-2.0 — commercial use allowed; use of the hosted Agent Scan API is subject to Snyk's terms.
Not really. It is a one-shot scanner; background and push-key modes exist for fleet use but are aimed at managed deployments.
Cisco AI Defense's open-source scanner for agent skills: YARA and YAML rules, dataflow analysis and an optional LLM judge, with SARIF output and a pre-commit hook.
NVIDIA's open-source security scanner for agent skills, with 71 detection patterns, optional LLM review and a skill-inspector skill.
It is a command-line scanner from Snyk that discovers agents, MCP servers and skills on your machine and checks them for prompt injection, sensitive-data handling issues and hidden malware payloads.
There is no npm package. Run it with uvx or download the standalone binary from GitHub Releases, and set SNYK_TOKEN to an API token from your Snyk account before scanning.
The tool is open source and its releases are signed, but a scan can start MCP servers listed in your configs and sends component content to Snyk. Run it in a sandbox when scanning configs you do not trust.
Agent Scan covers agents, MCP servers and skills and uses Snyk's hosted analysis API with a token. Cisco's skill-scanner focuses on skills; compare their rules and whether you can send skill content to a vendor service.