The Agent Skills Directory

[PROMPT_INJECTION]: Multiple reference files, such as references/failure-taxonomy.md, references/prompting-risks.md, and references/mistakes-security.md, contain well-known prompt injection and jailbreak strings (e.g., 'Ignore previous instructions', 'DAN', 'Do Anything Now'). A technical review confirms these are documented as examples within a security failure taxonomy and risk guide. They are provided as reference data for auditing purposes and are not intended for the agent to execute as instructions.
[REMOTE_CODE_EXECUTION]: The skill is comprised entirely of Markdown-based documentation and reference guides. It contains no scripts, binary executables, or instructions that trigger the download or execution of remote code. The installation guidance provided in the README is standard for the platform's skill management.
[DATA_EXFILTRATION]: There are no detected patterns for sensitive data access, credential harvesting (e.g., .env, .ssh, .aws), or unauthorized network operations. The skill's functionality is limited to information retrieval and prompting guidance.
[COMMAND_EXECUTION]: No shell commands, system calls, or subprocess executions are present within the skill's instruction set or reference library.

prompt-engineering