The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface through its knowledge ingestion workflow.
Ingestion points: The ingest operation processes data from external URLs, user-provided text, and local files found in ~/llm-wiki/raw/.
Boundary markers: The instructions do not define clear delimiters or "ignore embedded instructions" headers when saving or reading this content.
Capability inventory: The skill has the capability to read/write files across the wiki structure and execute shell commands for directory management (ls, mkdir).
Sanitization: There is no mention of sanitizing or escaping the content retrieved from external sources before it is written to the wiki or processed during the query and lint operations.
[COMMAND_EXECUTION]: The skill uses shell commands to manage its directory structure.
Evidence: It uses ls ~/llm-wiki/wiki/index.md for initialization checks and mkdir -p to create the organizational structure within the user's home directory. These operations are limited in scope and follow the intended purpose of the skill.
[EXTERNAL_DOWNLOADS]: The ingest operation involves interacting with external network resources.
Evidence: The skill includes logic to "crawl and save" content from URLs provided by the user. While this is a core feature for building a knowledge base, it involves fetching data from arbitrary external domains.

llm-wiki