The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines a workflow for ingesting external, untrusted data (URLs, PDFs, files) into a project's folder structure. This content is subsequently processed and synthesized by the agent, creating a surface for indirect prompt injection.
Ingestion points: External data enters the agent context through the ingest procedure described in the platform skill, which saves fetched content to the external-sources/ directory.
Boundary markers: The skill advocates for "STOP-gated" procedures (scoping gates in research and decision-confirmation gates in consolidation) that mandate user intervention before significant transitions, providing a layer of human review.
Capability inventory: The workflow guides the agent to perform network fetching (ingest), file system writes across the three-layer layout, and content synthesis.
Sanitization: No specific sanitization or filtering of external source content is mentioned before it is processed in the research phase.

knowledge-base