The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest and process a wide variety of untrusted project files, creating a surface for indirect prompt injection.
Ingestion points: The agent is directed in SKILL.md to scan all file types, including documentation (.md), diagrams (.svg, .drawio, .puml, .mmd), data files (.csv, .json, .yaml), and source code (comments and string literals).
Boundary markers: There are no boundary markers or instructions to treat ingested content as data only, increasing the risk that embedded instructions in processed files could influence agent behavior.
Capability inventory: The skill leverages the Edit tool with replace_all=true to modify files throughout the project directory.
Sanitization: No content validation or sanitization mechanisms are defined to filter or escape instructions embedded within the files being scanned.

french-language