build_from_json crashes on edges with 'from'/'to' keys instead of 'source'/'target'

[pedrolbacelar]

Bug

build_from_json in graphify/build.py crashes with KeyError: 'source' when edges use alternate key names like from/to instead of the expected source/target.

How to reproduce

When using LLM subagents (particularly smaller/faster models) for semantic extraction, they sometimes produce edges with {"from": "node_a", "to": "node_b", ...} instead of {"source": "node_a", "target": "node_b", ...}. This is a reasonable variation that the schema doesn't enforce at the subagent level.

When these edges reach build_from_json, line 45 does:

src, tgt = edge["source"], edge["target"]

This raises KeyError and crashes the entire graph build, losing all extraction work (including paid LLM tokens).

Expected behavior

build_from_json should normalize common alternate key names (from → source, to → target) before processing, and skip edges that are truly malformed (missing both variants).

Impact

• Edges with valid relationships are silently lost or cause crashes
• Smaller models (Sonnet, Haiku) are more likely to produce alternate key names
• The crash happens after the expensive LLM extraction, so the token cost is wasted

Suggested fix

for edge in extraction.get("edges", []):
    if "from" in edge and "source" not in edge:
        edge["source"] = edge.pop("from")
    if "to" in edge and "target" not in edge:
        edge["target"] = edge.pop("to")
    if "source" not in edge or "target" not in edge:
        continue
    src, tgt = edge["source"], edge["target"]

I have a PR ready for this: #217 (pending)

[safishamsi]

Fixed in v0.4.2. Edges using from/to keys are now normalized to source/target before processing, so they are no longer silently dropped.