The Agent Skills Directory

[PROMPT_INJECTION]: No attempts to override agent behavior, bypass safety guidelines, or extract system prompts were detected in the instructions or metadata.
[DATA_EXFILTRATION]: No sensitive file access (e.g., credentials, SSH keys) or network operations to external domains were identified. The skill operates locally on model configurations.
[REMOTE_CODE_EXECUTION]: No remote scripts are downloaded or executed. The provided Python snippets are static examples for model layer freezing and data structuring.
[COMMAND_EXECUTION]: The skill does not contain any shell commands, subprocess calls, or privilege escalation patterns like sudo or chmod.
[EXTERNAL_DOWNLOADS]: No external packages or dependencies are dynamically installed. References to third-party tools like Unsloth are purely informational.

vision-sft