Show HN: I scan AI agent skills for prompt injection before you install them
1 points
1 hour ago
| 1 comment
| github.com
| HN
adamos486
1 hour ago
[-]
I built Skulto because I was copy-pasting skills between Claude Code, Cursor, and Codex. Every agent has its own directory, no way to sync.

The security scanner is the part I'm most proud of. 35+ patterns detect prompt injection, jailbreaks, system prompt spoofing, shell injection (rm -rf, curl | sh), and Unicode obfuscation.

Also: symlink-based installs (one source of truth), offline-first, pure-Go binary, and an MCP server so Claude Code can search and install without leaving the terminal.

brew install asteroid-belt/tap/skulto

Happy to talk architecture or security patterns.

reply