The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest and analyze external data from "upstream repo" and "model card" documents (Phase 0). This represents an indirect prompt injection attack surface, as external repositories or model cards could contain malicious instructions designed to influence the agent's behavior during the integration process. The skill lacks specific boundary markers or sanitization instructions for this external content.
[COMMAND_EXECUTION]: The workflow involves the execution of various shell commands including pip install, uv run, git rm, and flashdreams-run. While these are standard tools for a developer environment, they provide a mechanism for local system interaction that requires appropriate oversight when executed by an autonomous agent.

integrate-a-model