The Agent Skills Directory

Indirect Prompt Injection (MEDIUM): The skill implements a migration strategy that ingests data from untrusted sources without sanitization or boundary markers.
Ingestion points: migrateFromMarkdown uses glob('**/*.md') to read all markdown files in the directory tree. migrateFromSQLite reads all entries from a local SQLite database.
Boundary markers: Absent. The skill reads file content and stores it directly into the vector index (this.agentdb.store) without delimiters or instructions to ignore embedded commands.
Capability inventory: The skill has file system read access (fs.readFile) and persistent storage capabilities via AgentDBAdapter. While no direct shell execution is shown, the indexed content influences the agent's semantic search and future reasoning.
Sanitization: Absent. There is no validation or filtering of the content being indexed.
Data Exposure (LOW): The migration logic uses a broad glob pattern (**/*.md) which may inadvertently ingest sensitive information contained in markdown notes or documentation files into the agent's memory backend.

V3 Memory Unification