The Agent Skills Directory

[PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection by ingesting untrusted data from external sources.
Ingestion points: The skill processes data from TASKS.md and various Model Context Protocol (MCP) tools including Chat, Email, Documents, and Calendar (SKILL.md steps 5 and 6).
Boundary markers: Absent; the instructions do not specify using delimiters or 'ignore' instructions when writing external content to memory files.
Capability inventory: The skill possesses file-writing capabilities to the local working directory to create and update memory files (SKILL.md step 7).
Sanitization: No explicit sanitization, validation, or filtering of the extracted content is described before it is written to the CLAUDE.md and memory/ files.

start