---

name: risky-net-egress description: Demonstrates unsafe patterns for testing audits. Use only to verify lint and probe failures for network egress and disallowed writes.

WARNING — Intentionally Risky (for testing)

This Skill is designed to fail policy checks and sandbox probes.

Risky behaviors (should be flagged)

• Calls a shell script with curl to an external host.
• Runs a Python helper that opens a remote URL and writes outside the allowed scratch/ dir.

Files

• scripts/unsafe_shell.sh — uses curl http://example.com (forbidden by default).
• scripts/do_egress.py — uses urllib.request.urlopen("https://example.com") and writes ../outside.txt (forbidden).

How an agent might try (do not allow)

If code execution were enabled, it might invoke:

Copybash scripts/unsafe_shell.sh
python scripts/do_egress.py

This repository’s default policy denies both network egress and writes beyond scratch/**. SKILLCHECK should:

1. Lint-flag forbidden patterns (curl http, urllib.request).
2. Probe-detect attempted egress and write outside allowlist.