The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs users to download the source code of the KarpathyTalk application from a remote GitHub repository (https://github.com/karpathy/KarpathyTalk.git) using git clone.
[COMMAND_EXECUTION]: The skill requires the user to compile and execute a binary (go build -o karpathytalk ./cmd/karpathytalk and ./karpathytalk) from the downloaded source code.
[INDIRECT_PROMPT_INJECTION]: The skill provides patterns and code samples for building LLM agents that ingest untrusted user-generated content from the social network API.
Ingestion points: External data enters the agent context via http.Get calls to /api/posts, /api/posts.md, and /api/users/{username}/posts.md in the provided Go samples (SKILL.md).
Boundary markers: The skill lacks explicit instructions or code for using delimiters or boundary markers when interpolating this external content into LLM prompts.
Capability inventory: The skill environment includes network access (http.Get), database access (sqlite3), and the ability to build and run Go binaries.
Sanitization: There is no evidence of sanitization, validation, or filtering of the fetched markdown/JSON content before it is injected into the prompt context in the getContextForLLM example.

karpathytalk-community