The Agent Skills Directory

[SAFE]: No malicious patterns, obfuscation, or unauthorized data access techniques were found across the skill's documentation and reference files.
[PROMPT_INJECTION]: The skill provides standard prompt engineering patterns (such as role-playing, XML delimiters, and few-shot examples) to steer agent behavior. These are instructional and do not attempt to bypass model safety filters or extract sensitive system information.
[COMMAND_EXECUTION]: Potentially dangerous commands like rm -rf and git push --force are mentioned only within a 'Confirm Before Risky Actions' section. This is a security-positive instruction intended to prevent accidental or autonomous damage by the agent.
[DATA_EXFILTRATION]: Instructions regarding file access are focused on reducing hallucinations by ensuring the agent reads relevant code before answering. There are no patterns suggesting exfiltration of sensitive data or hardcoded credentials.
[REMOTE_CODE_EXECUTION]: The skill contains no instructions to download or execute external scripts from untrusted sources. Code snippets in the documentation are for illustrative API configuration purposes only.

prompting-agents