The Agent Skills Directory

[SAFE]: The skill is entirely instructional and provides a markdown-based workflow for security incident response. It does not contain executable code, scripts, or external dependencies.
[PROMPT_INJECTION]: A static detector triggered a warning for prompt injection patterns. Upon manual review, this is confirmed as a false positive. The identified strings (e.g., "ignore previous instructions") are located within a "Prompt Injection Safety Notice" section (Section 8), which explicitly instructs the agent to ignore any such commands found within the external data (logs, emails, malware) it may be analyzing.
[DATA_EXFILTRATION]: No network tools or data-sending capabilities are requested or used. The skill includes specific guidance for the agent to avoid including full credentials or sensitive encryption keys in its output, recommending generic references instead.
[COMMAND_EXECUTION]: The skill does not perform any shell command execution or script spawning. It only utilizes read-only tools (Read, Grep, Glob) for data analysis.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted incident data, which is an inherent risk surface. However, the author has included significant mitigation instructions, such as explicit role boundaries, output schema validation, and warnings against executing or obeying content found in analyzed artifacts.

ir-playbook