skillsmith helps people inspect AI-agent skill files before they install them. The free browser tool combines structural checks and static security signals; the accompanying guides explain the practical risks in plain language. The project is operated by Lars Hänggi in Switzerland and its scanner code is publicly available for inspection.
Our guides focus on a narrow question: how text-based instructions can change an agent’s behavior, access sensitive data, or introduce supply-chain risk. Each guide is written to help a reader make a concrete decision: what to inspect, what a warning signal means, and when to stop and investigate further. The scanner, API documentation, FAQ, and legal pages are maintained separately so that product instructions do not masquerade as editorial advice.
skillsmith’s scanner is static-first: it examines submitted text and metadata rather than executing a skill on a user’s machine. Its checks cover malformed frontmatter, instruction-override language, suspicious outbound destinations, credential-access hints, download-and-run patterns, obfuscation, and known vulnerable dependencies. Where package versions are available, the scanner can query the public OSV vulnerability database.
Detection patterns are maintained in the public codebase. Changes are testable and reviewable through the repository history. Read the full detection-methodology guide alongside the underlying skill; neither replaces careful review.
Security tools reduce uncertainty; they do not remove it. A static scan cannot establish an author’s intent, detect every new phrase, or safely run untrusted code. For high-impact skills, read the file, review dependencies, apply least privilege, and use an isolated environment. Behavioral analysis is an additional heuristic aid, not a security guarantee.
For product feedback, false positives, corrections, or accessibility issues, open an issue in the public GitHub repository. Legal identity and address are on the Impressum; data-handling details are in the Privacy Policy.