The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted operational evidence and timeline signals, which presents a surface for indirect prompt injection if the external data sources contain malicious instructions.
Ingestion points: Untrusted data enters the agent context via the 'Inputs To Gather' section in SKILL.md (timeline signals, operational evidence) and is processed into assets/incident-timeline-template.csv and assets/postmortem-report-template.md.
Boundary markers: The templates and workflow instructions do not include explicit delimiters or instructions to ignore embedded prompts within the evidence being analyzed.
Capability inventory: The skill is designed to read and write files within the skill's directory to update reports, timelines, and trackers.
Sanitization: No evidence of sanitization, escaping, or validation logic for the external input was found in the provided instructions or templates.

incident-postmortem