The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted binary data from .ase and .aseprite files which can contain arbitrary strings in fields such as layer names, tag names, and user-defined properties. This represents an indirect prompt injection surface as an agent processing the output might interpret these strings as instructions.
Ingestion points: The scripts/aseprite-inspect.mjs script uses readFileSync to load user-provided binary files for parsing.
Boundary markers: Extracted strings are placed directly into the JSON output without specific delimiters or warnings to the consuming agent to ignore embedded instructions.
Capability inventory: The skill's capabilities are limited to reading the input file and writing JSON to standard output. It does not have access to the network, the ability to write files, or the ability to execute arbitrary shell commands.
Sanitization: Content is parsed from binary chunks using the standard TextDecoder and included in the output without filtering for potential prompt injection patterns.
[SAFE]: The skill implements proactive resource management to mitigate potential denial-of-service attacks. The inspectAseprite function in scripts/_lib/asset-tools.mjs uses a maxDecompressMib limit (default 64 MiB) when calling zlib.inflateSync to prevent decompression bomb attacks (zip bombs) from exhausting system memory.
[SAFE]: The parser implementation includes rigorous bounds checking. The Reader class in scripts/_lib/asset-tools.mjs verifies available buffer space using need() and take() methods before every read operation, ensuring the script does not attempt to read past the end of the file data.

aseprite