[Parked pending adopter demand] UnderstandingFrame coding-agent context artifact

Context

Generated code is only the projection; the durable asset is the validated understanding around assumptions, boundaries, prior failures, and evidence.

ContextWeaver is a good place to make that idea operational: when a coding agent receives context, it should not only receive files. It should also receive relevant lessons, known assumptions, contracts, and risks.

Goal

Extend context assembly so coding agents can receive an explicit "understanding frame" alongside source snippets.

Proposed behavior

When assembling context for a coding task, ContextWeaver should be able to include:

• relevant prior lessons;
• known data/schema assumptions;
• boundary conditions;
• previous failure modes;
• validation requirements;
• open risks or uncertainty notes.

Example scenario

Task: modify a pandas/spreadsheet ingestion flow.

ContextWeaver should be able to inject something like:

Known assumptions:
- input spreadsheets are expected to contain stable column headings
- numeric columns may arrive as strings

Prior lessons:
- validate headers and dtypes before numeric coercion
- fail fast with a schema error before transformation

Validation requirements:
- include a malformed-header fixture
- include a non-numeric value fixture

Acceptance criteria

• Define a small UnderstandingFrame or equivalent context section.
• Add retrieval hooks for prior lessons / assumptions, even if initially file-backed or stubbed.
• Include this section in rendered context when relevant items exist.
• Add an example for dataframe/spreadsheet transformation context.
• Tests verify that relevant lessons and assumptions are included without replacing normal source context.

Non-goals

• Do not build the lesson registry inside ContextWeaver.
• Do not require a specific LLM provider.
• Do not force every context package to include an understanding frame.

No files or tests are named; start by locating the context-assembly entry point and existing rendering tests. Define the frame and its retrieval boundary, then verify that relevant lessons and assumptions are rendered alongside normal source context, including the dataframe/spreadsheet example, without requiring every package to include it.