The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill enables agents to store and recall shared knowledge using the nitro agent memory command. This creates an attack surface for indirect prompt injection, where instructions stored in memory by one session could influence the behavior of an agent in a future session.
Ingestion points: Content is ingested from the shared store via context, search, recent, and show subcommands in SKILL.md.
Boundary markers: While the skill suggests storing discrete facts, it does not specify the use of delimiters when content is placed into the prompt context.
Capability inventory: The skill allows for local command execution via the nitro CLI and subsequent task-related tools.
Sanitization: The instructions do not define a process for sanitizing or validating memory content before it is processed by the agent.
[COMMAND_EXECUTION]: The skill uses the nitro CLI to perform memory operations. The instructions correctly direct the user to the vendor's official installation page at chillicream.com rather than attempting automated installation, which is a safe practice.

nitro-memory