The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run within its wrapper scripts (scripts/wrapper-template.py and the auto-generated wrappers) to invoke the local parser script. The implementation uses a list of arguments rather than a shell string, which is a recommended security practice to prevent shell injection vulnerabilities.
[DATA_EXFILTRATION]: The script scripts/whatsapp-group-summary.py reads WhatsApp message logs stored locally in ~/.hermes/whatsapp/messages.jsonl. While these logs contain sensitive communication data, the skill processes this information locally to generate a summary for the user. No evidence of unauthorized network transmission to external or untrusted domains was found.
[PROMPT_INJECTION]: Since the skill processes untrusted chat data (messages from a WhatsApp group), it is inherently exposed to indirect prompt injection. However, the skill includes a sanitize_text function in scripts/whatsapp-group-summary.py specifically designed to strip out invisible Unicode characters and control codes that are frequently used to obfuscate malicious instructions in prompt injection attacks.
[DYNAMIC_EXECUTION]: The scripts/create-group-wrapper.py script is used to generate local Python wrapper scripts from a template. This is a standard configuration step for the skill's functionality and does not involve downloading remote code or executing untrusted dynamic content.
[CREDENTIALS_UNSAFE]: The documentation in SKILL.md guides the user on how to configure their own phone numbers and access policies. It encourages the use of an allowlist policy for Direct Messages, which enhances the security of the resulting bot deployment.

hermes-wag-summarizer