RAG-grounded code generation — retrieves attune context and emits answers with source citations

Failure modes

Symptom	Cause	Fix	Severity
`RuntimeWarning: coroutine 'RagCodeGenWorkflow.execute' was never awaited`	`execute` called without `await`	It is a coroutine — `await` it or use `asyncio.run`	high
`error` is `"query argument is required"`	`execute` called with an empty/missing `query`	Pass a non-empty `query`	high
`RuntimeError: ... needs the attune-rag package ...`	`attune-rag` (a core dependency) is not installed	`pip install attune-rag`	high
`error` is `"RAG retrieval failed: ..."`	The pipeline raised (corpus I/O, connection, timeout, bad variant)	Check corpus availability / connectivity; retry	medium
`error` is `"k argument must be an integer ..."`	`k` wasn't an int (e.g. `k="bad"`)	Pass an integer `k`	low
`error` is `"unknown model ..."`	`model` isn't in `MODEL_REGISTRY`	Use a registered model id, or omit `model`	low
`DeprecationWarning` about `cwd=`	Passing the deprecated `cwd` alias	Use `path=` instead	low

The async call is easy to get wrong. execute is a coroutine; forgetting to await it is the most common mistake.
attune-rag must be installed. It's a core dependency, not an optional extra — the workflow can't retrieve without it.
Slug vs. feature name. The CLI slug is rag-code-gen; the feature/help topic is rag-grounding.