The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: A curl | bash pattern was detected in templates/security_guard_selftest.py. Analysis confirms this is part of a test suite used to verify that the security guard permits standard installers while blocking malicious variants; it is not executed by the skill for non-testing purposes.
[COMMAND_EXECUTION]: The skill uses subprocess.run to interact with system tools for its intended functions. runtime_footer.py executes curl to fetch balance information from an internal vendor-provided API (starchild-credit-api.internal), while self-test scripts use subprocesses to validate hook logic.
[PROMPT_INJECTION]: The skill implements context injection via pre_llm_call hooks, such as in runtime_footer.py which injects a directive to prevent the model from fabricating cost data. This is an intended feature of the hook architecture used for behavioral steering.
[DATA_EXFILTRATION]: The skill includes a security_guard.py template specifically designed to detect and block data exfiltration attempts, such as attempts to pipe sensitive files like .env or SSH keys to remote network tools.

agent-hooks