Skip to content

Give an agent a map of its codebase

uncoded indexes Python code and Markdown docs. It gives the agent progressively deeper views of Python code:

  1. A map shows the code symbols
  2. Stubs add parameters and types
  3. uncoded body returns exact source on demand.

The agent can stop as soon as one view gives it enough information.

Get started How agents use it

What the agent sees

For a project containing this function, uncoded generates a map with its name and location.

The source

src/greetings.py
def greet(name: str) -> str:
    return f"Hello, {name}!"

The map

.uncoded/namespace.yaml (excerpt)
src/:
  greetings.py:
    greet:

The map is the first view. It shows that greet exists and where to find it. When the agent needs the function's parameters and return type, it reads the generated stub:

.uncoded/stubs/src/greetings.pyi (excerpt)
def greet(name: str) -> str: ...

When the agent needs the exact source, it retrieves that definition without reading the rest of the source file:

uv run uncoded body greet --in src/greetings.py

Broad questions may need only the map. Questions about names and types may stop at the stub.

Before a rename, uncoded refs finds the references to check and update. Markdown docs get a separate map of files and headings.

The index stays local. The installed skill files are committed with the project so agents know how to use it.