Function run_jcode

scripts/compare_token_usage.py:164–241

Run jcode and capture token usage from trace output.

(prompt: str, workdir: str, jcode_binary: str, model: str = "claude-opus-4-5-20251101")

def run_jcode(prompt: str, workdir: str, jcode_binary: str, model: str = "claude-opus-4-5-20251101") -> RunResult:
    """Run jcode and capture token usage from trace output."""
    try:
        # Create a temporary JCODE_HOME to avoid polluting user's sessions
        with tempfile.TemporaryDirectory() as tmpdir:
            env = os.environ.copy()
            env["JCODE_HOME"] = tmpdir
            env["JCODE_TRACE"] = "1"

            result = subprocess.run(
                [
                    jcode_binary,
                    "run",
                    "--no-update",
                    "--model", model,
                    prompt,
                ],
                capture_output=True,
                text=True,
                cwd=workdir,
                timeout=120,
                env=env,
            )

            # Parse token usage from trace output in stderr
            # Format: [trace] token_usage input=X output=Y cache_read=Z cache_write=W
            input_tokens = 0
            output_tokens = 0
            cache_read = 0
            cache_write = 0

            for line in result.stderr.split("\n"):
                if "[trace] token_usage" in line:
                    parts = line.split()
                    for part in parts:
                        if part.startswith("input="):
                            input_tokens = int(part.split("=")[1])
                        elif part.startswith("output="):
                            output_tokens = int(part.split("=")[1])
                        elif part.startswith("cache_read="):
                            cache_read = int(part.split("=")[1])
                        elif part.startswith("cache_write="):
                            cache_write = int(part.split("=")[1])

            token_usage = TokenUsage(
                input_tokens=input_tokens,
                output_tokens=output_tokens,
                cache_read_tokens=cache_read,
                cache_creation_tokens=cache_write,
            )

            return RunResult(
                tool="jcode",
                prompt=prompt,
                usage=token_usage,
                success=result.returncode == 0,
                output=result.stdout,
                error=None if result.returncode == 0 else result.stderr,