The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill architecture establishes a persistent memory system where learned patterns and task results are stored in ChromaDB collections. This creates a surface for indirect prompt injection where data processed in one session can be retrieved and injected into the agent's context in future sessions.
Ingestion points: Data enters the memory system via the storeImprovement function (documents) and is retrieved via retrieveRelevantImprovements (query texts) in SKILL.md.
Boundary markers: The implementation does not demonstrate the use of boundary markers or 'ignore instructions' warnings when interpolating retrieved memory into the agent's prompt.
Capability inventory: The skill utilizes mcp__chroma__* tools for database operations and loadAgentConfig for local file access.
Sanitization: The provided logic does not include sanitization or validation of the stored insights before they are re-injected into the agent's working context.
[COMMAND_EXECUTION]: The skill documentation describes a system that relies on executing tool commands for managing vector database collections and reading local configuration files from the /agents/ directory.

agent-memory-skills