The Agent Skills Directory

[PROMPT_INJECTION]: The SKILL.md file includes instructions designed to override agent behavior, such as directives to "Ignore Claude-specific mode-switch instructions" and a "Strict execution contract" requiring the agent to execute the skill protocol exactly as written. These meta-instructions are intended to prioritize the skill's specific rules over the agent's general safety and operational guidelines.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external DOCX files, which creates an indirect prompt injection surface where malicious instructions embedded in a document could influence the agent when it processes the resulting Markdown.
Ingestion points: DOCX files are ingested via the --input argument and processed in scripts/lib/converter.cjs.
Boundary markers: The conversion output does not include boundary markers or warnings to the agent that the converted content is untrusted data.
Capability inventory: The skill utilizes file system write capabilities (fs.writeFileSync) in scripts/lib/converter.cjs and scripts/lib/output-handler.cjs to save Markdown and extracted images.
Sanitization: While the skill uses mammoth and turndown for conversion, it does not perform sanitization of the resulting text content to filter for prompt injection patterns.

docx-to-markdown