apodex/frontier-agent-demo

1"""Unicode math-symbol sanitization for Python code blocks."""2from __future__ import annotations3 4_UNICODE_REPLACEMENTS: dict[str, str] = {5    "\u222B": "integral",   # Integral sign6    "\u2211": "sum",        # Summation7    "\u221E": "inf",        # Infinity8    "\u00B2": "**2",        # Superscript 29    "\u00B3": "**3",        # Superscript 310    "\u2074": "**4",        # Superscript 411    "\u207B": "-",          # Superscript minus12    "\u00B9": "**1",        # Superscript 113    "\u27E8": "<",          # Left angle bracket14    "\u27E9": ">",          # Right angle bracket15    "\u2329": "<",          # Left-pointing angle bracket16    "\u232A": ">",          # Right-pointing angle bracket17    "\u2014": "-",          # Em dash18    "\u2013": "-",          # En dash19    "\u2018": "'",          # Left single quote20    "\u2019": "'",          # Right single quote21    "\u201C": '"',          # Left double quote22    "\u201D": '"',          # Right double quote23    "\u2264": "<=",         # Less than or equal24    "\u2265": ">=",         # Greater than or equal25    "\u2260": "!=",         # Not equal26    "\u00D7": "*",          # Multiplication sign27    "\u00F7": "/",          # Division sign28    "\u2248": "==",         # Almost equal29    "\u2245": "==",         # Approximately equal30    "\u2261": "==",         # Identical to31    "\u2192": "->",         # Right arrow32    "\u2190": "<-",         # Left arrow33    "\u221A": "sqrt",       # Square root34    "\u03C0": "pi",         # Pi35}36 37 38def sanitize_code(code: str) -> str:39    """Replace Unicode math symbols with ASCII equivalents and strip non-ASCII40    bytes from comments.41 42    Also removes lines that start with a shell/jupyter escape (``!pip install…``)43    — those get fed to the Python interpreter, which rejects them with a44    ``SyntaxError``.45 46    Code-line behaviour (non-comment):47      * any char present in the replacement table is mapped to its ASCII form48      * non-ASCII chars that have no replacement are left alone — they may49        be legitimate (e.g. a unicode string literal)50 51    Comment-line behaviour:52      * symbol replacement runs first53      * any remaining non-ASCII bytes are replaced by ``?`` (lossy) — comments54        never affect execution, so we drop noise eagerly55    """56    if not code:57        return code58 59    out_lines: list[str] = []60    for line in code.split("\n"):61        # Strip jupyter/shell escapes.62        if line.lstrip().startswith("!"):63            continue64 65        if any(ord(ch) > 127 for ch in line):66            stripped = line.lstrip()67            if stripped.startswith("#"):68                for sym, repl in _UNICODE_REPLACEMENTS.items():69                    line = line.replace(sym, repl)70                # Lossy: comments may be in any locale, we don't need them71                # round-tripped.72                line = line.encode("ascii", "replace").decode("ascii")73            else:74                for sym, repl in _UNICODE_REPLACEMENTS.items():75                    line = line.replace(sym, repl)76        out_lines.append(line)77 78    return "\n".join(out_lines)79