The Agent Skills Directory

[PROMPT_INJECTION]: No prompt injection or safety bypass instructions were found in the skill content or metadata. The instructions focus entirely on technical test development.
[DATA_EXFILTRATION]: No evidence of sensitive data access or unauthorized network exfiltration. The provided code templates use localized testing URLs (e.g., DEFAULT_URL_FOR_TEST).
[OBFUSCATION]: The text is clear and direct. No encoded commands, hidden characters, or homoglyph-based obfuscation were detected.
[REMOTE_CODE_EXECUTION]: No remote code execution patterns or downloads from untrusted sources were found. All references to external scripts and tools are part of the sgl-project's own repository structure.
[DYNAMIC_EXECUTION]: The skill documents the use of subprocesses via popen_launch_server to facilitate end-to-end testing of the inference server. This is a standard and safe practice within a development and CI context.
[COMMAND_EXECUTION]: The skill describes standard command-line usage for running test suites and checking configurations, with no malicious command patterns or unauthorized privilege escalation.

write-sglang-test