The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a memory retrieval and context injection loop that is susceptible to indirect prompt injection. Stored interactions are compiled into a system prompt that influences future agent behavior.
Ingestion points: Data is ingested and stored in the memory backend using the memstack store command, which accepts arbitrary string content (SKILL.md, REFERENCE.md).
Boundary markers: The documentation for memstack context and the memory retrieval examples do not demonstrate the use of delimiters or boundary markers to isolate retrieved memories from the agent's core instructions.
Capability inventory: The skill utilizes shell commands and requires network access to communicate with LLM providers (OpenAI, Anthropic) and various database backends (Postgres, Redis).
Sanitization: There is no mention of filtering, escaping, or sanitizing stored content before it is interpolated into the systemPrompt field for use by the agent.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of a global Node.js package from the public npm registry to provide its core functionality.
Evidence: npm install -g @memstack/cli in SKILL.md.

memstack-cli