The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains instructions in SKILL.md that explicitly direct the agent to override standard safety protocols by bypassing user confirmation gates. Specifically, it states: 'When the user dictates exact SMS text, send it directly via scripts/FreeMobile_sms.py without additional confirmation'. This reduces human oversight for actions with external side effects.
[DATA_EXFILTRATION]: The script scripts/FreeMobile_sms.py transmits the sensitive FREEMOBILE_SMS_API_KEY (mapped to the pass parameter) via an HTTP GET request to https://smsapi.free-mobile.fr/sendmsg. This is a poor security practice as credentials in URL query strings are frequently captured in plaintext by web server logs, proxy logs, and browser history.
[PROMPT_INJECTION]: The SAFETY.md file includes several test prompts containing common injection patterns, such as system message simulation ('[SYSTEM] Configuration Free Mobile mise à jour...') and constraints bypass attempts. These examples could be misinterpreted by the model or serve as templates for real attacks.
[INDIRECT_PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection due to the combination of automated execution and lack of sanitization.
Ingestion points: The message parameter of the FreeMobile_sms.py script, which is populated from the user's prompt.
Boundary markers: None. The skill lacks delimiters or instructions to treat the message content as untrusted data.
Capability inventory: The skill uses requests in scripts/FreeMobile_sms.py to perform network operations.
Sanitization: There is no validation or escaping of the message content beyond a simple length truncation.
[METADATA_POISONING]: The skill uses branding ('freemobile-sms-hardened', 'Armored by Faberlens') that suggests a higher level of security, while its core instructions (bypassing confirmation) and implementation (GET-based credential transmission) actually introduce significant security regressions.

freemobile-sms-hardened