The Agent Skills Directory

[SAFE]: The skill implements robust resource limiting in scripts/common.py through MAX_INPUT_BYTES and MAX_STDIN_BYTES constants. This protects the agent's environment from memory-based Denial of Service (DoS) attacks when processing excessively large or crafted inputs.
[SAFE]: The safe_write_bytes function in scripts/common.py follows secure coding patterns by writing to a temporary file and using atomic renaming (os.replace). It also includes an explicit check to refuse writing through symlinks, which prevents path traversal or symlink redirection attacks targeting sensitive system files.
[SAFE]: The skill utilizes a binary sniffing mechanism (looks_binary) that identifies common file signatures for ZIP/Office containers, PDFs, and images. This prevents the deterministic scripts from accidentally mangling binary data by attempting to decode it as UTF-8 text.
[SAFE]: The instructions in SKILL.md and references/chinese-text.md establish strict boundaries for the agent, explicitly requiring the preservation of technical identifiers, code blocks, and legal disclosures. This reduces the risk of functional data loss or unintended modification of structured content.
[SAFE]: The stylometric analysis in scripts/score_stylometry.py and Unicode cleaning in scripts/text_unicode.py use standard libraries and deterministic logic. No dynamic code execution (eval, exec) or unsafe deserialization of user-provided data was found.

text-watermark-cleaner-zh-tw