Potential risk: normalize_direction() silently remaps unknown aliases to 'forward' with no warning (commit f71bad5)

[kapil971390]

Bug Report

Commit: f71bad5 — fix(skills/understand): persist canonical edge direction during merge

---

What was fixed

The PR correctly canonicalizes edge direction values (both → bidirectional, case normalization, invalid → forward) before merging graphs, fixing duplicate edges leaking into the final knowledge-graph.json.

Remaining risk

normalize_direction() silently converts any unknown value to forward with no warning:

_DIRECTION_ALIASES = {"both": "bidirectional", "mutual": "bidirectional"}
_VALID_DIRECTIONS  = {"forward", "backward", "bidirectional"}

def normalize_direction(value):
    candidate = str(value).lower().strip() if value else "forward"
    candidate = _DIRECTION_ALIASES.get(candidate, candidate)
    return candidate if candidate in _VALID_DIRECTIONS else "forward"  # silent fallback

If a batch graph ships an edge with a new valid alias (e.g., "reverse" or "both-ways") that is not yet in _DIRECTION_ALIASES, it silently becomes "forward" — changing the semantic meaning of the edge with no log entry.

Impact scenario

A future LLM-generated batch may emit "direction": "reverse" (a reasonable alias for "backward"). With the current code:

• "reverse" is not in _DIRECTION_ALIASES
• "reverse" is not in _VALID_DIRECTIONS
• Edge silently becomes "forward" — opposite direction
• The knowledge graph ships with inverted relationships, no error raised

Suggested fix

def normalize_direction(value):
    candidate = str(value).lower().strip() if value else "forward"
    candidate = _DIRECTION_ALIASES.get(candidate, candidate)
    if candidate not in _VALID_DIRECTIONS:
        logger.warning(f"Unknown edge direction {value!r} — defaulting to forward")
        return "forward"
    return candidate

A warning here costs nothing and makes data quality issues visible during merges.