The Agent Skills Directory

[COMMAND_EXECUTION]: Explicit instructions to bypass system-level network security controls.
Evidence: The skill documentation instructs the agent to use env -u HTTPS_PROXY -u HTTP_PROXY and curl --noproxy '*' to intentionally circumvent sandbox proxy restrictions.
Impact: This allows the agent to establish unmonitored network connections, bypassing organizational security filters and traffic logging.
[CREDENTIALS_UNSAFE]: Sensitive account credentials are exported into the shell environment.
Evidence: The script fetches and exports Pushover login emails and passwords using export PO_EMAIL and export PO_PW.
Impact: Placing raw credentials in environment variables makes them accessible to any subprocess and potentially exposes them in process listings or system logs.
[EXTERNAL_DOWNLOADS]: The skill facilitates the download and processing of arbitrary remote content.
Evidence: The helper script make_custom_sound.sh accepts a <url> parameter to fetch MP3 files from external sources.
Impact: Processing untrusted remote files through media processing pipelines (likely FFmpeg) without explicitly documented sanitization creates an attack surface for media-based exploit vectors.
[DATA_EXFILTRATION]: Capability to extract and transmit sensitive API tokens outside of monitored channels.
Evidence: The WEB create-app command is designed to capture and reveal 30-character API tokens.
Impact: Combined with the proxy bypass instructions, this capability allows a malicious actor to exfiltrate newly generated access tokens to an external endpoint without detection.

manage-apps-and-sounds-headless