The Agent Skills Directory

Indirect Prompt Injection Surface: The skill establishes a mechanism where the agent reads and applies context or definitions from workspace documentation files.
Ingestion points: Data enters the agent's context through files like CLAUDE.md, memory/glossary.md, and markdown files located within the memory/people/, memory/projects/, and memory/context/ directories.
Boundary markers: The instructions do not define strict boundary markers or include explicit warnings to isolate or ignore administrative/natural language instructions that might be present within the memory records.
Capability inventory: The skill enables the agent to interpret these records to fulfill user requests, which can involve tasks such as composing messages or analyzing workflows with the retrieved context.
Sanitization: There is no mention of input filtering, escaping, or structural validation for the data ingested from these memory documents, representing a potential layout for indirect prompt interaction if files are modified by external inputs or untrusted contributors.

memory-management