The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external web sources to ground its simulations in real-world intelligence, which creates an inherent attack surface for indirect prompt injection.
Ingestion points: The skill utilizes multiple parallel searches via WebSearch and WebFetch in Step 2.5 and Step 3 to gather data from external websites, articles, and reports.
Boundary markers: The subagent prompts lack explicit delimiters or specific instructions to the models to disregard instructions potentially hidden in the fetched external content.
Capability inventory: The skill uses Agent tool calls to spawn parallel sub-processes and performs local file writes to create simulation dossiers and update the project's MEMORY.md index.
Sanitization: No explicit sanitization or filtering of the content retrieved from external URLs is mentioned before the data is incorporated into the intelligence briefing and processed by the subagents.

doctor-strange