The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes a dedicated safety section instructing the agent to treat all alert data as passive evidence rather than executable instructions. It explicitly warns against following directives found within log payloads, effectively mitigating potential injection attempts targeting the analyst workflow.
[INDIRECT_PROMPT_INJECTION]: As a tool designed to ingest untrusted log data, this skill has an inherent attack surface. However, it implements robust defensive measures, including role boundary enforcement and output validation requirements, which reduce the risk of indirect prompt injection to a negligible level.
[DATA_EXFILTRATION]: The skill instructions proactively advise on data protection by requiring the redaction of sensitive credentials and tokens from any generated reports. No unauthorized network activity or data transmission patterns were detected.
[COMMAND_EXECUTION]: The skill defines clear role boundaries, specifying that it is intended for analysis and recommendation only, not for direct system modification or command execution. The allowed tools (Read, Grep, Glob) are appropriate for its stated purpose.

alert-triage