The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents a workflow for feeding source code chunks directly into an LLM context, which is a known vector for indirect prompt injection if the code contains malicious instructions. * Ingestion points: File content is ingested via the ts-pack process command and the process(source_code, config) SDK function (SKILL.md). * Boundary markers: There are no instructions for using delimiters or protective prompting to isolate the code chunks from the agent's instructions. * Capability inventory: The skill uses the ts-pack CLI and tree-sitter-language-pack library to read and manipulate file data (SKILL.md). * Sanitization: No sanitization or escaping mechanisms are described for the content being sent to the LLM.

chunking-for-llms