templates/skills/sandbox/SKILL.md
Execute JavaScript, shell commands, fetch web content, and control a browser
npx skillsauth add ddalcu/agent-orcha sandboxInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
Run JavaScript in isolated VM. Use console.log() for output, return for result. No fetch/require — use sandbox_web_fetch for HTTP.
Run shell commands in /tmp with limited permissions (Docker only).
Fetch web pages (auto-converted to markdown) or APIs (raw: true).
Search DuckDuckGo. Params: query, num_results.
Observe → Act → Observe loop. Elements have short refs (e1, e2...) — use them in click/type.
| Tool | Use |
|------|-----|
| sandbox_browser_observe | Text snapshot with element refs. Primary verification tool. |
| sandbox_browser_navigate | Go to URL, wait for ready, return snapshot |
| sandbox_browser_click | Click by ref ("e3") or text ("Submit") |
| sandbox_browser_type | Type by ref ("e5"), dispatches input/change events |
| sandbox_browser_content | Page as markdown (optional selector) |
| sandbox_browser_evaluate | Run JS, reports side effects |
| sandbox_browser_screenshot | PNG screenshot (expensive, last resort) |
sandbox_web_fetch for HTTP — not fetch in sandbox_execsideEffects in evaluate resultstesting
A test skill for unit testing
databases
A skill that requires sandbox
tools
Browser automation — observe-act loop with refs
testing
PII and OWASP security filtering rules