The Agent Skills Directory

[DATA_EXFILTRATION]: The skill is designed exclusively for local-only operation. A review of src/memory.py confirms that no network libraries (such as requests, urllib, or socket) are imported or used. All data is persisted in a local SQLite database stored in the user's home directory.
[COMMAND_EXECUTION]: The codebase does not utilize any subprocess management, shell execution, or dynamic code evaluation functions. The skill's operations are limited to standard Python data structures and database interactions.
[PROMPT_INJECTION]: As a memory system, the skill ingests data that could originate from untrusted sources. The mandatory evidence chain analysis shows:
Ingestion points: Data enters through the remember, learn, and track_entity methods in src/memory.py.
Boundary markers: The skill's integration instructions in SKILL.md explicitly mandate user review of all recalled data before the agent performs any consequential actions.
Capability inventory: The skill is restricted to local file I/O (SQLite) with no access to system commands or networking.
Sanitization: All database interactions are implemented using parameterized SQL queries to prevent injection vulnerabilities.
[CREDENTIALS_UNSAFE]: No hardcoded credentials, API keys, or tokens were detected. The skill includes explicit documentation and guardrails instructing agents and users against storing sensitive information in the memory database.

agent-memory-hardened