The Agent Skills Directory

[SAFE]: The skill implements a local storage solution for agent 'memories' using a SQLite database located at ~/.mem/mem.db. Technical analysis confirms that all data handling is local, with no network-facing code, unauthorized data exfiltration, or hardcoded credentials detected.
[EXTERNAL_DOWNLOADS]: The skill uses date-fns and nanoid from the official NPM registry. These are well-known, versioned libraries used for standard date formatting and ID generation. No unverifiable remote scripts or automated downloads from untrusted sources are present.
[COMMAND_EXECUTION]: The CLI tool manages a local database but does not utilize exec(), eval(), or subprocess calls to run arbitrary shell commands. It uses the bun:sqlite library with prepared statements for all database interactions, preventing SQL injection.
[DATA_EXFILTRATION]: No network capabilities (such as fetch, curl, or HTTP requests) are present in the source code. Local file access is limited to reading user-specified images for storage and writing exported images to a user-defined directory.
[PROMPT_INJECTION]: The skill stores untrusted data (user notes or stdin) which are later retrieved into the agent's context. While this creates an attack surface for indirect prompt injection:
Ingestion points: Data enters via stdin or command-line arguments in src/cli.ts.
Boundary markers: None are present in the stored data.
Capability inventory: The skill only performs local file read/write and SQLite operations; it cannot execute shell commands or access the internet.
Sanitization: All inputs are handled using prepared SQL statements to prevent injection into the database engine. Given the limited capabilities of the skill, this risk is considered low.

mem