AI Agent Frameworks

Browse prompts in this category.

Prompts

AI Agent Framework Security Architect
Design agent apps on LangChain, CrewAI, AutoGen, LlamaIndex, Semantic Kernel, and Claude Agent SDK — tool schemas, credential isolation, sandboxed code executors, delegation boundaries, HITL gates, input/output guardrails. Use when choosing framework-level defaults for an agent app.
AutoGen Secure Agent Development
Harden AutoGen/AG2 agents — DockerCommandLineCodeExecutor over local, max_consecutive_auto_reply limits, inter-agent injection filtering, register_for_execution/llm separation, human_input_mode discipline, masked logging. Use when building or reviewing AutoGen conversations and group chats.
Claude Agent SDK Secure Development
Harden Claude Agent SDK apps — JSON Schema tool validation, sandboxed bash/computer-use with no ambient credentials, supervisor handoffs, human approval for destructive ops, prompt guardrails, scoped API keys with cost controls. Use when building or reviewing Claude-powered agents.
CrewAI Secure Multi-Agent Development
Harden CrewAI crews — scoped agent roles/goals, restricted allow_delegation, Pydantic tool schemas, human_input on destructive tasks, per-crew memory isolation with TTLs, max_rpm/max_iter controls, and expected_output enforcement. Use when building or auditing CrewAI multi-agent crews.
LangChain and LangGraph Secure Development
Harden LangChain and LangGraph — Pydantic args_schema tools, typed LangGraph state with interrupt_before checkpoints, persistent SQL checkpointing, PythonREPLTool sandboxing, LangSmith masking, per-user RAG filtering. Use when building or reviewing LangChain chains, graphs, or callbacks.
LlamaIndex Secure RAG Agent Development
Harden LlamaIndex RAG agents — IngestionPipeline scanning, per-tenant indexes with MetadataFilters, similarity thresholds, FunctionTool Pydantic schemas, tree_summarize synthesis, and callback PII masking. Use when building or reviewing LlamaIndex retrieval, query engines, or agents.