The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/message_guard.py includes a run_tests() function that utilizes subprocess.run() to invoke the script's own --gate mode for behavioral verification. This execution is limited to the local Python interpreter and the skill's own source file.
[DYNAMIC_EXECUTION]: The test script scripts/test_message_guard.py uses importlib.util to dynamically load the message_guard.py module. This is a standard development practice for unit testing local scripts and does not involve loading untrusted external code.
[SAFE]: The skill implements a fail-closed security mechanism for agent correspondence. It requires no external network connectivity, contains no hardcoded credentials, and relies exclusively on Python standard libraries. It accesses local configuration files (e.g., ~/.claude/settings.json) solely to verify that its security hooks are correctly wired.

synthesis-message-guard