The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) because it ingests and processes untrusted data from Word (.docx) files.
Ingestion points: Document content is extracted from the file path provided in $ARGUMENTS in SKILL.md.
Boundary markers: Absent. The instructions do not direct the agent to use delimiters or 'ignore embedded instructions' warnings when presenting the extracted document content.
Capability inventory: The agent has access to Bash (used for script execution and dependency installation) and the Read tool (used for displaying images).
Sanitization: The extract_docx.py script extracts raw text and formulas without sanitization, which the agent then presents directly in its context.
[COMMAND_EXECUTION]: The SKILL.md file contains instructions to execute a Python script using a hardcoded, absolute Windows file path (D:/BaiduSyncdisk/claude_code/student_info/.claude/skills/read-word/scripts/extract_docx.py). While not a direct security threat, this reveals information about the author's local environment and will lead to execution failures on any system that does not have an identical directory structure.

read-word