The Agent Skills Directory

[SAFE]: The skill implements robust file handling security by using os.replace for atomic writes and verifying that paths are not symlinks before writing, which prevents symlink-following attacks.\n- [SAFE]: It enforces strict input limits (256 MiB for files and 64 MiB for stdin) and subprocess resource caps (4 GiB address space, 2 GiB file size) to mitigate resource exhaustion and Denial of Service (DoS) risks.\n- [SAFE]: The code includes comprehensive logic for detecting and removing invisible Unicode carriers, zero-width characters, and homoglyphs, effectively neutralizing potential text-based obfuscation or steganography.\n- [SAFE]: The instructions provide clear boundaries for the AI agent, explicitly forbidding the modification of code, identifiers, APIs, and protected technical spans, which reduces the risk of accidental corruption or indirect injection vulnerabilities during text processing.\n- [SAFE]: The Python scripts are self-contained, utilize the standard library, and do not make any network requests or download external dependencies, ensuring a closed and verifiable execution environment.

clean-user-facing-text