The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external Claude skill files (SKILL.md and reference files) and possesses shell execution capabilities, creating a surface for indirect prompt injection.
Ingestion points: The skill reads source files from ~/.claude/skills/ or a user-provided path in Step 1.
Capability inventory: The skill executes shell commands including node, npm install, and rm in Step 8 and 9.
Boundary markers: There are no explicit markers or instructions to ignore embedded prompts within the source skill data being processed.
Sanitization: No sanitization or validation of the input skill content is mentioned before it is processed into the JSON data for document generation.
[COMMAND_EXECUTION]: The conversion workflow involves executing shell commands to manage dependencies, run the document generator, and clean up temporary files.
Evidence: Step 8 instructs the agent to run node to execute the generator and potentially npm install to satisfy dependencies. Step 9 uses rm to delete the temporary JSON file.
[EXTERNAL_DOWNLOADS]: Fetches the docx library from the npm registry to support document generation.
Evidence: Step 8 includes a fallback instruction to run npm install docx in ~/Documents if the package is missing.
[DYNAMIC_EXECUTION]: The skill follows a dynamic execution pattern where it generates a temporary JSON configuration file based on analyzed content and then invokes a Node.js script to process that data.
Evidence: Step 7 involves writing a JSON file ([skill-name]-conversion.json) which is then passed as an argument to docx-generator.js in Step 8.

skill-to-prompt