The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted strings that are included in reports, creating a surface for indirect prompt injection.
Ingestion points: Untrusted data enters the agent context through methods such as Incident.add_action, IncidentTimeline.add_event, EvidenceTracker.add_item, LessonsLearned.set_summary, and LessonsLearned.add_finding within the scripts/ir_utils.py file.
Boundary markers: There are no explicit delimiters or instructions defined in the skill to treat these ingested strings as untrusted data or to ignore embedded instructions.
Capability inventory: The skill includes the ability to write data to the local filesystem using the IncidentTimeline.export_csv method in scripts/ir_utils.py.
Sanitization: No validation, escaping, or sanitization is performed on the ingested strings before they are incorporated into generated reports or exported files.

incident-response