The Agent Skills Directory

[DYNAMIC_EXECUTION]: The Python scripts scripts/alignment_score.py and scripts/synthetic_bitext_recipes.py calculate the path to the internal _linguistic_shared directory and add it to the module search path via sys.path.insert to load bundled dependencies. This is a local configuration for module resolution within the skill package.
[INDIRECT_PROMPT_INJECTION]: The skill workflow involves processing external linguistic datasets (parallel bitext and monolingual corpora). While this presents a potential attack surface for indirect prompt injection if the source data contains malicious instructions, the skill implementation focuses on data alignment and filtering.
Ingestion points: Parallel and monolingual corpora files (SKILL.md, Step 2 & 5).
Boundary markers: None explicitly mentioned in the instructions for input data.
Capability inventory: The provided scripts perform metadata analysis and recommendation generation; they do not include dangerous operations like arbitrary code execution or network writes.
Sanitization: No explicit text sanitization or filtering for injection patterns in source data is specified in the workflow.
[EXTERNAL_DOWNLOADS]: The skill documentation and reference files (e.g., references/mining_recipes.md, references/canonical_sources.md) point to reputable linguistic resources and research repositories, such as those from SIL International, Meta Research, and the OPUS project.

magic-linguistic-bitext