The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires the installation of the yomitan-tools package from the NPM registry to function.
[INDIRECT_PROMPT_INJECTION]: The skill processes and renders dictionary entries from external files, creating a potential surface for indirect prompt injection attacks.
Ingestion points: Dictionary data is loaded from local .zip files or directories via the openDictionary method and accessed through various search and read functions.
Boundary markers: The instructions do not specify any delimiters or safety prompts to ensure the agent ignores instructions embedded within dictionary glossary or tag fields.
Capability inventory: The skill allows the agent to read file content and persist modifications to the file system through save() and saveAs() methods.
Sanitization: The render method converts entries to Markdown for review, but there is no documentation regarding the sanitization or escaping of potentially malicious instructions hidden in the dictionary data.

yomitan-tools