The Agent Skills Directory

[COMMAND_EXECUTION] (SAFE): The provided scripts (cipher_tools.py, steganography.py, spectrogram.py) perform mathematical transformations and image/audio processing locally. They do not spawn subprocesses, execute shell commands, or use dynamic execution functions like eval() or exec().
[DATA_EXFILTRATION] (SAFE): No network operations or sensitive file access patterns were found. File operations are limited to reading/writing image, audio, and text files as specified by the user via command-line arguments.
[EXTERNAL_DOWNLOADS] (SAFE): The skill does not perform any runtime downloads of scripts or binaries. The documentation references well-known external tools (e.g., CyberChef, dCode), but these are provided as informational links only.
[OBFUSCATION] (SAFE): While the skill includes tools to create and extract hidden messages (steganography and zero-width Unicode encoding), these are the intended functions of the toolkit for puzzle design. The skill's own instructions and code are clear and not obfuscated.
[PROMPT_INJECTION] (SAFE): The markdown files contain instructional content and reference material for ARG design. No patterns were detected that attempt to override the AI agent's system prompt or bypass safety filters.
[Indirect Prompt Injection] (SAFE):
Ingestion points: scripts/steganography.py and scripts/spectrogram.py process external data (images, audio, and text) to extract hidden content.
Boundary markers: Absent in the extraction scripts.
Capability inventory: Limited to local file read/write and data transformation. No network or system-level capabilities are available to be exploited by extracted data.
Sanitization: Absent, but the lack of dangerous capabilities mitigates the risk.

arg-designer