The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted external text data (large-scale monolingual corpora) which presents a surface for indirect prompt injection.
Ingestion points: Untrusted content enters the agent's context when identifying candidate corpora (Step 1), performing paragraph-level Language Identification (Step 3), and calculating deduplication statistics (Step 5).
Boundary markers: The instructions rely on paragraph-level processing but do not explicitly mandate the use of XML tags or secure delimiters when the agent handles or summarizes the content of these corpora.
Capability inventory: The skill has the capability to write to the file system (via interaction_utils.safe_json_dump) and execute local Python scripts included in the package.
Sanitization: The workflow includes mandatory Unicode NFC normalization and confusable character folding (Step 4) to mitigate orthographic manipulation and improve data quality.
[COMMAND_EXECUTION]: The Python scripts in the scripts/ directory dynamically modify sys.path to include a relative sibling directory (_linguistic_shared). While this is a form of dynamic loading from computed paths, the computation is based on the script's own file location (__file__) and does not incorporate user-supplied input, making it a common and relatively safe implementation pattern for self-contained skill packages.

magic-linguistic-corpus