The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from security logs and alerts (such as process command lines, DNS queries, and alert reasons) which could be manipulated by an attacker to influence the agent's classification or decision-making process.
Ingestion points: Data is fetched from .alerts-security.alerts-*, logs-endpoint.events.process-*, and logs-endpoint.events.network-* indices in SKILL.md (Steps 1, 3, and 4).
Boundary markers: The skill instructs the agent to "report tool output verbatim" and "distinguish facts from inference," but does not define explicit delimiters or sanitization steps for the ingested data.
Capability inventory: The agent has the capability to execute Elasticsearch queries (elastic es search, elastic es query) and update alert statuses via the Detection Engine API (elastic kb security-detections-api set-alerts-status).
Sanitization: No specific filtering or escaping of raw log content is performed before the agent evaluates the data for threat classification.
[SAFE]: The skill utilizes the elastic CLI, a tool provided by the author (Elastic), to perform intended security operations. The operations are well-scoped to the task of security alert investigation, and the classification logic in references/classification-guide.md emphasizes evidence-based analysis.

security-alert-analysis