The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary functionality involves learning from user input to steer future agent behavior (e.g., goals, actions, routing). This creates a surface for indirect prompt injection where an attacker could influence the agent's long-term behavior by providing instructions that become memorized patterns. Ingestion points: The memory_recall and memory_learn tools process user-provided input_text and value fields. Boundary markers: The skill implements a memory-blind baseline and requires a specific, one-time confirmation phrase for any changes suggested by memory. Capability inventory: The skill performs database writes and executes CLI commands for its memory operations. Sanitization: Instructions require filtering out credentials and high-entropy secrets before storage.
[COMMAND_EXECUTION]: The instructions in SKILL.md and references/recall-and-apply.md explicitly command the agent to run local Python scripts (scripts/memory.py, scripts/validate_skills.py) via the python3 command for core functionality.
[DYNAMIC_EXECUTION]: The scripts/mcp_server.py script uses sys.path.insert to dynamically load the memory module from the local file system at runtime.

nerd-memory