The Agent Skills Directory

[PROMPT_INJECTION]: No evidence of instructions intended to bypass safety filters, override system prompts, or extract internal instructions was found.
[DATA_EXPOSURE]: The skill does not access sensitive local files (like SSH keys or AWS credentials) and does not hardcode any authentication secrets. It uses standard environment variable patterns for configuration.
[REMOTE_CODE_EXECUTION]: No patterns of downloading and executing scripts from untrusted remote sources were detected.
[COMMAND_EXECUTION]: The skill uses SDK-specific registration methods for application logic but does not invoke arbitrary shell commands or unsafely interpolate user input into system-level operations.
[OBFUSCATION]: The content is clear and technical. No hidden characters, encoded payloads (Base64, hex), or homoglyph attacks were identified.
[PRIVILEGE_ESCALATION]: No attempts to gain administrative or root privileges, or to modify system-level permissions, were found.
[PERSISTENCE]: The skill does not attempt to create startup scripts, scheduled tasks, or other mechanisms to maintain access across sessions.
[INDIRECT_PROMPT_INJECTION]: While the skill processes external event data, it relies on structured JSON filters and engine-level schema validation (args_format, request_format) to define and validate operations, following best practices for handling untrusted data.

iii-observe