The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill architecture is designed to ingest and process data from external services, APIs, and multiple AI models to provide cascading responses. This pattern creates a potential vulnerability surface for indirect prompt injection if external sources provide adversarial content.
Ingestion points: The skill ingests data from external APIs (e.g., weather data, search results) and model outputs (e.g., GPT-4/GPT-3.5 analysis) across primary, secondary, and tertiary fallback levels (SKILL.md).
Boundary markers: The instructions do not mandate or provide examples of using strict delimiters or "ignore previous instructions" safety prompts when interpolating external data into the agent's context.
Capability inventory: The workflow utilizes the store_discovery function from amplihack.memory.discoveries to persist results and metrics to memory, providing a mechanism where injected data could potentially influence future sessions (SKILL.md, Step 7).
Sanitization: There are no explicit instructions for validating, filtering, or sanitizing external content before it is used for decision-making or stored in memory.

cascade-workflow