The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface for indirect prompt injection through its text processing and consistency scanning features.
Ingestion points: In references/zh2en.md and references/en2zh.md, the agent is instructed to read content from the user-provided academic text and all previously translated .md files located in the {Project Root Path}/translation/ directory.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands (e.g., 'ignore previous instructions' within the text) are provided in the workflow logic.
Capability inventory: The skill relies on the Read, Write, and Edit tools to manipulate local files, and uses the Python tool to run helper scripts.
Sanitization: There is no evidence of sanitization or filtering of instructions contained within the external files before they are processed by the agent.
[COMMAND_EXECUTION]: The skill instructs the agent to execute local Python scripts included in the package (scripts/validate-glossary.py and scripts/extract-terms.py) to perform term extraction and glossary validation. While the scripts themselves are benign, they represent the execution of local code shipped with the skill.
[DATA_EXPOSURE]: The skill accesses a hidden path in the user's home directory (~/.claude/academic-translate/glossary.md) to maintain a global terminology list. Although this is used for persistence of translation terms, it involves reading and writing to a hidden system path.

academic-translate