The Agent Skills Directory

[PROMPT_INJECTION]: The skill facilitates the creation of agents that ingest untrusted user input through prompt variables (e.g., payload.get("prompt", "") in references/python-patterns.md and agent.invoke(prompt) in references/typescript-patterns.md). This input is processed by agents that may have access to high-privilege tools, presenting an indirect prompt injection surface.
Ingestion Points: User-provided prompts are received in the invoke entrypoints defined in references/python-patterns.md and references/typescript-patterns.md.
Boundary Markers: No explicit delimiters (like XML tags) or system instructions are provided in the templates to differentiate between user input and developer instructions.
Capability Inventory: The framework includes a pre-built bash tool (@strands-agents/sdk/vended-tools/bash) that allows for arbitrary shell command execution.
Sanitization: The provided code patterns do not include logic for sanitizing or validating user input before passing it to the agent's reasoning engine.
[COMMAND_EXECUTION]: The TypeScript patterns in references/typescript-patterns.md explicitly demonstrate the use of a bash tool. While this is a documented feature for 'DevOps assistant' use cases, it grants the agent the ability to execute code on the host system where the agent is deployed.

strands-agent