agentsop-prompt-history-inspect

To inspect the rendered prompt during LLM debugging, you must dump the wire-level input sent to the provider. This skill enforces a 30-second inspection step to reveal prompt divergence before you edit source code or retry generation.

Source prompts differ from wire-level inputs because framework render layers transform your code before transport. This skill provides cross-framework visibility for DSPy, LangChain, LangGraph, CrewAI, and Aider to classify whether divergence originates in code, framework, or provider.

Yes, this prompt inspection method works with raw OpenAI and Anthropic SDK calls. It provides a unified SOP and logging mapping that covers raw SDKs alongside higher-level frameworks to capture the exact LLM input transmitted.

The best way to debug production-only LLM failures and token cost spikes is to inspect the rendered prompt first. This skill helps you capture production traces and classify the root cause of latency, truncation, refusals, or schema violations across frameworks.

To trace prompt rendering divergence before changing LLM generation settings, dump the rendered prompt history immediately upon seeing surprising output. This skill enforces a pre-edit inspection step to provide evidence-oriented troubleshooting for framework rendering layers.

Yes, this prompt debugging approach works with CrewAI and LangGraph workflows. It delivers cross-framework prompt visibility and unified commands to inspect the actual LLM input, resolving schema violations and format failures across these specific frameworks.