[Bug]: SQL engine flags all '--' line comments as injection — false positives erode boundary trust

[rahuldass19]

QWED-MCP Audit Finding: LOW-03

Severity: 🟢 LOW
Component: sql_engine.py — INJECTION_PATTERNS
Audit ID: LOW-03
Re-audit of v0.2.1: Finding is NEW

---

Description

The injection pattern set includes:

r"--[^\n]*",

This flags every SQL line comment as a potential injection — including idiomatic, documented queries:

SELECT id, email FROM users  -- only fetch non-PII columns
WHERE active = 1

verify_sql_query returns verified=False for this query. Any developer schema that ships with inline comments (standard practice in migration files and ORM-generated SQL) is blocked wholesale.

False positives at the verification boundary are not harmless: repeated blocking of legitimate code trains consumers to distrust or bypass the guard — the same erosion dynamic documented in #13 for open().

---

Why This Violates QWED Philosophy

• Principle 3 — Deterministic Decisions: A comment is not deterministic evidence of injection. The pattern encodes a heuristic ("comments are suspicious") that produces both noise and, conversely, no protection against actual comment-based bypasses in this architecture (the query never executes here — the engine only judges patterns).
• Consistency with [Bug]: Code engine blocks open() causing false positives on legitimate QWED SDK usage #13's lesson: guards that block safe code while missing dangerous alternatives are worse than no guard.

---

Expected Behavior

• Remove the bare --[^\n]* pattern.
• Retain genuinely dangerous comment-adjacent patterns only where they combine with injection structure (e.g., ' ; -- statement termination followed by comment is already covered by '\s*;\s*--).
• Regression test: a commented legitimate query passes; ' OR '1'='1' -- still fails.

---

Environment

Field	Value
QWED-MCP Version	0.2.1
Component	SQL Verification Engine
File	src/qwed_mcp/engines/sql_engine.py (L29–38)