The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches installation scripts and binaries from ollama.com, which is the official domain for the well-known Ollama service.
[REMOTE_CODE_EXECUTION]: Utilizes the standard Linux installation procedure for Ollama by piping a remote script to the shell; this behavior is consistent with the official installation guidelines for the tool.
[COMMAND_EXECUTION]: Employs system commands (e.g., uname, free, nvidia-smi) to assess hardware capabilities and manages local services via systemd, Homebrew, or Docker, which is necessary for the skill's primary function.
[CREDENTIALS_UNSAFE]: Includes a placeholder secret (sk-prod-abc123xyz) within a mock git diff in the examples documentation; this is clearly used for educational purposes to demonstrate code review functionality and does not represent an actual credential exposure.
[PROMPT_INJECTION]: Features a code review example that processes untrusted git diff data (Ingestion: reviewDiff function in references/examples.md). The implementation uses backticks and specific system instructions as boundary markers, restricts capabilities to build status reporting (process.exit(1)), and applies JSON parsing for output sanitization to mitigate risks of indirect prompt injection.

ollama-setup