The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external incident data to generate postmortem reports, which creates a vulnerability surface for indirect prompt injection.
Ingestion points: The script scripts/generate_postmortem.py reads JSON incident data from files or standard input.
Boundary markers: The generated Markdown output does not utilize delimiters or specific instructions to the agent to ignore embedded commands within the incident data.
Capability inventory: The skill includes a Python script capable of performing file system read and write operations.
Sanitization: While the script validates specific enum-like fields such as severity and event types, free-text fields (e.g., descriptions, root cause analysis, action items) are rendered directly into the output without escaping or sanitization.

incident-retrospective