The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the iris CLI to provision the Redis memory service, create API keys, and perform status checks. It correctly employs flags such as --no-input and --json for machine automation and recommends using --write-env to store sensitive credentials in a local environment file rather than displaying them in the terminal output.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates the extraction of long-term memories from user conversations, which introduces a potential surface for indirect prompt injection.
Ingestion points: Conversations are processed through the redis_iris library (e.g., via convo.user() and convo.assistant()) to record and recall facts.
Boundary markers: Memories are retrieved and injected into the agent's system prompt using the as_prompt() method, which formats the data for the model.
Capability inventory: The skill includes shell command execution via the iris CLI and file system operations, including writing to service.yaml and .env files.
Sanitization: The configuration allows for a specific exclusions.prompt to be defined in the service.yaml file, providing natural language instructions to the underlying service to ignore and avoid storing sensitive information like passwords or recovery codes.

add-agent-memory