The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides explicit shell command templates in references/web-clipper-usage.md to invoke an external utility (web-clipper). These templates use arguments such as URLs and file paths that may be derived from untrusted user input or search results, creating a surface for command injection vulnerabilities.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and interpret data from untrusted sources, which serves as a major attack surface for indirect prompt injection.
Ingestion points: In SKILL.md, the workflow includes recursive scanning of local directories (Step 0.2) and fetching content from official websites and search results (Step 1).
Boundary markers: The skill does not provide any boundary markers or instructions to the agent to differentiate between its core operating instructions and potential malicious instructions embedded within the source materials (PDFs, websites, etc.).
Capability inventory: The agent has the capability to execute shell scripts, perform recursive file system reads, and write structured markdown files to build the output repository.
Sanitization: There is no documentation or instruction regarding the sanitization or validation of content before it is processed by the LLM for extraction or writing-style analysis.
[EXTERNAL_DOWNLOADS]: The skill orchestrates the automated retrieval of remote content from arbitrary web domains to populate the knowledge base. While it uses an external tool for this purpose, the skill manages the logic of where and how these downloads occur.

enterprise-clone-builder