The Agent Skills Directory

[PROMPT_INJECTION]: The skill facilitates indirect prompt injection by processing untrusted content from remote URLs and local files.
Ingestion points: The skill accepts URLs, local file paths, and folder paths as input for translation.
Boundary markers: While the skill includes instructions to preserve formatting and placeholders, it lacks explicit security boundaries to prevent the agent from executing instructions embedded within the source text.
Capability inventory: The skill has the ability to perform network reads (fetching content from URLs) and filesystem writes (saving translated files to the workspace root).
Sanitization: Validation steps are limited to translation accuracy and formatting consistency, with no security sanitization of the input text.
[DATA_EXFILTRATION]: The skill initiates network requests to fetch content from user-provided URLs.
The skill uses network operations to retrieve data from external, non-whitelisted domains provided by the user.
The skill explicitly lists .env files as a supported format for translation, which involves reading files commonly used to store sensitive credentials.

tech-translator