The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill integrates external content (FAQs and Skills) into the agent's prompt, creating a potential surface for indirect prompt injection if the content source is compromised.\n
Ingestion points: Content is ingested from local files in faqs/ and skills/ directories and retrieved at runtime from the Redis Iris service via the redis-iris SDK.\n
Boundary markers: The skill suggests rebuilding context from the base prompt each turn to prevent instruction leakage between turns, though it does not mandate specific delimiters for the injected content.\n
Capability inventory: The agent is granted capabilities to discover playbooks and load procedural instructions using registered tools (load_skill, discover_playbook).\n
Sanitization: Security relies on a 'content authoring' workflow where content is validated, planned, and published by a human or trusted CI process before being served to the agent.\n- [COMMAND_EXECUTION]: The skill provides instructions for using the iris CLI tool to manage agent playbooks, including provisioning services, managing keys, and synchronizing content.\n
Evidence: Multiple iris agent-playbooks commands are documented for administration, extraction binding, and content publication.\n
Context: These are standard administrative operations for the Redis Iris platform and include safety flags like --json --no-input and --dry-run.\n- [DYNAMIC_EXECUTION]: The skill uses dynamic instruction loading where the agent identifies and loads 'Skills' (procedures) at runtime based on the conversation context.\n
Mechanism: The load_skill tool allows the agent to fetch specific procedural instructions that are not part of the initial system prompt.\n
Evidence: Documentation in references/integration.md describes tool registration for progressive disclosure and score-based hooks.

add-agent-playbooks