The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill clones external git repositories based on user-provided URLs or paths. This involves downloading content from potentially untrusted remote sources.
[COMMAND_EXECUTION]: The skill uses terminal commands including git clone and git ls-remote to interact with external repositories. It also references an ExecutionProxy interface (using the rtk prefix) for executing terminal actions such as npm test during project scaffolding.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external repositories to scaffold agent instructions and configurations, which provides a surface for embedded instructions in third-party code.
Ingestion points: External git repositories provided by the user (SKILL.md, references/from-repo-guide.md).
Boundary markers: No explicit delimiters or instructions to ignore embedded prompts within the cloned content are provided.
Capability inventory: File system writing, git operations, and terminal execution via the ExecutionProxy.
Sanitization: There is no mention of sanitization or validation of the content discovered in the external repositories before it is incorporated into the generated agent packages.

company-creator