Slipstream

Slipstream is the "people do not say the whole thing" project.

It used to be called Undercurrent. The problem is still the same: humans compress intent because that is normal, and software has to either fill the gaps well or waste everyone's time asking obvious questions.

What It Does

Slipstream tries to enrich a request before an agent acts on it:

• classify what the person is probably asking for
• pull context from the conversation, files, or git
• notice the assumptions
• ask only when a missing answer actually changes the path
• hand the agent a better starting point

The Pipeline

The SDK breaks that into four quiet stages:

1. Classify the intent, specificity, scope, and emotional load.
2. Harvest context from adapters such as conversation history, the filesystem, and git.
3. Analyze the gaps: what can be inferred, what needs a stated assumption, and what truly needs a question.
4. Compose the enriched prompt so the agent starts with the missing context already attached.

The reference version is deterministic and dependency-light. The LLM strategy can drop in later, but the useful thing is the shape: context first, questions last.

Why I Keep Coming Back To It

Bad prompts are usually not lazy. They are compressed. A good system should understand that and meet the person halfway.

Current Shape

Slipstream is no longer just a note in the margin. It is a shipped TypeScript SDK with adapters, middleware, hooks, tests, and the same rule this workspace follows: never make the user fill out a questionnaire when the system can infer the answer.

The Bar

If it feels like a questionnaire, it failed. If it feels like the software quietly understood what was between the lines, it worked.