Useful bias manipulation re: LLM – the stochastic parrot speaks

The source of this conversation came from working with Claude on an MCP project for providing semi-autonomous guidance to an LLM while working on non-trivial coding tasks.

It was an interesting conversation that I had the presence of mind to have Claude preserve.

It grew from having to constantly refocus the LLM away from presumption of the target audience being a User rather than an LLM instance. The modus-operandi.xml doc was initially an attempt to cram the directions and and cues useful for keeping it on-track during mcp development while using a local copy of the prototype knowledge store.