Function fetch_catalog_via_jcode

scripts/benchmark_discovery.py:357–418

(
    args: argparse.Namespace,
    socket_path: Path,
    workdir: Path,
    categories: list[str],
)

def fetch_catalog_via_jcode(
    args: argparse.Namespace,
    socket_path: Path,
    workdir: Path,
    categories: list[str],
) -> dict[str, list[dict[str, Any]]]:
    created = json.loads(
        run_debug_command(args, socket_path, f"create_session:{workdir}", timeout=60)
    )
    session_id = created["session_id"]
    catalog: dict[str, list[dict[str, Any]]] = {}
    try:
        for index, category in enumerate(categories, start=1):
            last_error: Exception | None = None
            for attempt in range(1, args.catalog_retries + 1):
                tool_input = json.dumps(
                    {
                        "category": category,
                        "query": (
                            f"public {category.replace('-', ' ')} capabilities available "
                            "for external agent workflows"
                        ),
                        "reason": (
                            "Enumerate the live public listings so benchmark scenarios can be "
                            "validated without selecting or configuring any provider."
                        ),
                    },
                    separators=(",", ":"),
                )
                try:
                    raw = run_debug_command(
                        args,
                        socket_path,
                        "tool",
                        f"discover_tools {tool_input}",
                        session_id=session_id,
                        timeout=20,
                    )
                    payload = json.loads(raw)
                    call = parse_discovery_output(str(payload.get("output", "")), 0.0)
                    if call.category != category or call.outcome not in {"listing", "empty"}:
                        raise BenchmarkError(
                            f"unexpected catalog response for {category}: {payload.get('output', '')}"
                        )
                    catalog[category] = [{"name": name} for name in call.tools]
                    break
                except Exception as error:
                    last_error = error
                    if attempt < args.catalog_retries:
                        time.sleep(args.retry_delay * attempt)
            else:
                raise BenchmarkError(
                    f"catalog discovery failed for {category} after "
                    f"{args.catalog_retries} attempts: {last_error}"
                )
            progress(index, len(categories), "categories", f"Fetched catalog category {category}")
    finally:
        try: