The Agent Skills Directory

[COMMAND_EXECUTION]: The add function in scripts/sab-api.sh is vulnerable to code injection. It interpolates the user-provided URL directly into a python3 -c command string using bash variable expansion: python3 -c "import urllib.parse; print(urllib.parse.quote('$url', safe=''))". An attacker can provide a URL containing a single quote followed by Python code to execute arbitrary commands with the permissions of the agent process.
[CREDENTIALS_UNSAFE]: In scripts/sab-api.sh, the api_call function appends the SAB_API_KEY directly to the URL as a query parameter (apikey=${SAB_API_KEY}). This is an insecure practice because query parameters are often recorded in cleartext in server logs, proxy logs, and shell history.
[DATA_EXFILTRATION]: The add-file command in scripts/sab-api.sh uses curl -F "nzbfile=@$filepath". The @ prefix instructs curl to read the file at the specified path. If a user or a malicious indirect source tricks the agent into providing a path to sensitive files (such as ~/.ssh/id_rsa or .env), the content of those files will be sent to the configured SABnzbd server.
[PROMPT_INJECTION]: The SKILL.md file contains aggressive instruction hijacking language ("⚠️ MANDATORY SKILL INVOCATION ⚠️") intended to override the agent's decision-making process and force the use of the skill regardless of the agent's safety or operational assessments.

sabnzbd