Ngrok Preview

Generate short-lived, mobile-friendly ngrok preview links for local artifacts and share them in Telegram. Use when OpenClaw produces images/charts/generated...

This skill appears to do what it says: create short-lived ngrok previews of files you explicitly provide. Before installing/using it: 1) Install ngrok from the official site and provide NGROK_AUTHTOKEN securely if you want authenticated tunnels. 2) Only pass specific artifact paths (not broad or system dirs) — the script will copy whatever paths you give into ~/.cache/openclaw-ngrok-preview, so do not include secrets (SSH keys, config dirs, root of workspace). 3) Run the provided cleanup/down commands after the preview is no longer needed to remove local copies. 4) Verify the preview URL before sharing (it exposes files over the internet). If you want automated Telegram posting, note the skill does not contain Telegram integration or request a Telegram token — sharing must be done by the agent/user via their existing channel.

Type: OpenClaw Skill Name: ngrok-preview Version: 1.0.2 The 'ngrok-preview' skill is designed to expose local files and directories via ngrok, which is an inherently high-risk capability. While the SKILL.md explicitly instructs the AI agent to only publish 'task-specific outputs' and 'never broad directories', the underlying `scripts/ngrok_preview.py` script's `copy_sources` function will copy and serve any path provided to its `--source` argument. This creates a significant prompt injection vulnerability, allowing a malicious user to instruct the AI agent to expose sensitive local files (e.g., `~/.ssh`, `/etc/passwd`) or entire directories, leading to data exfiltration. The script uses `subprocess.Popen` to run a local HTTP server and the ngrok client, providing the mechanism for this exposure.