Compare commits
1 Commits
| Author | SHA1 | Date | |
|---|---|---|---|
|
3ce1f829a2 |
@@ -29,8 +29,6 @@ import logging
|
||||
import os
|
||||
import ssl
|
||||
import threading
|
||||
import time
|
||||
import uuid
|
||||
from http.server import BaseHTTPRequestHandler, HTTPServer
|
||||
from pathlib import Path
|
||||
from typing import Any, Callable, Dict, Optional
|
||||
@@ -443,244 +441,3 @@ class A2AMTLSClient:
|
||||
def post(self, url: str, json: Optional[Dict[str, Any]] = None, **kwargs: Any) -> Dict[str, Any]:
|
||||
data = (__import__("json").dumps(json).encode() if json is not None else None)
|
||||
return self._request("POST", url, data=data, **kwargs)
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# Structured A2A task delegation over mTLS
|
||||
# ---------------------------------------------------------------------------
|
||||
|
||||
_TERMINAL_TASK_STATES = {"completed", "failed", "canceled", "rejected"}
|
||||
|
||||
|
||||
def _iso_now() -> str:
|
||||
return time.strftime("%Y-%m-%dT%H:%M:%SZ", time.gmtime())
|
||||
|
||||
|
||||
def _task_status(state: str, message: str) -> Dict[str, Any]:
|
||||
return {
|
||||
"state": state,
|
||||
"message": message,
|
||||
"timestamp": _iso_now(),
|
||||
}
|
||||
|
||||
|
||||
def _coerce_artifact(result: Any) -> Dict[str, Any]:
|
||||
if isinstance(result, dict):
|
||||
if "text" in result:
|
||||
return result
|
||||
if "artifact" in result and isinstance(result["artifact"], dict):
|
||||
return result["artifact"]
|
||||
return {"text": str(result)}
|
||||
|
||||
|
||||
def _build_task_record(task_id: str, task: str, requester: Optional[str], metadata: Optional[Dict[str, Any]] = None) -> Dict[str, Any]:
|
||||
return {
|
||||
"taskId": task_id,
|
||||
"task": task,
|
||||
"requester": requester,
|
||||
"metadata": metadata or {},
|
||||
"artifacts": [],
|
||||
"status": _task_status("submitted", "Task submitted"),
|
||||
}
|
||||
|
||||
|
||||
def _default_agent_card(host: str, port: int) -> Dict[str, Any]:
|
||||
base_url = f"https://{host}:{port}"
|
||||
try:
|
||||
from agent.agent_card import build_agent_card
|
||||
from dataclasses import asdict
|
||||
|
||||
card = asdict(build_agent_card())
|
||||
except Exception as exc: # pragma: no cover - fallback only exercised when card build breaks
|
||||
logger.warning("Falling back to minimal agent card: %s", exc)
|
||||
card = {
|
||||
"name": os.environ.get("HERMES_AGENT_NAME", "hermes"),
|
||||
"description": "Hermes A2A task server",
|
||||
"version": "unknown",
|
||||
}
|
||||
card["url"] = base_url
|
||||
card["a2aTaskEndpoint"] = f"{base_url}/a2a/rpc"
|
||||
return card
|
||||
|
||||
|
||||
def _default_local_hermes_executor(task_payload: Dict[str, Any]) -> Dict[str, Any]:
|
||||
task_text = str(task_payload.get("task", "")).strip()
|
||||
if not task_text:
|
||||
return {"text": ""}
|
||||
from run_agent import AIAgent
|
||||
|
||||
agent = AIAgent(quiet_mode=True)
|
||||
result = agent.chat(task_text)
|
||||
return {
|
||||
"text": result,
|
||||
"metadata": {"executor": "local-hermes"},
|
||||
}
|
||||
|
||||
|
||||
class A2ATaskServer:
|
||||
"""JSON-RPC A2A task server running over the routing mTLS server."""
|
||||
|
||||
def __init__(
|
||||
self,
|
||||
cert: str | Path,
|
||||
key: str | Path,
|
||||
ca: str | Path,
|
||||
host: str = "127.0.0.1",
|
||||
port: int = 9443,
|
||||
executor: Optional[Callable[[Dict[str, Any]], Dict[str, Any]]] = None,
|
||||
card_factory: Optional[Callable[[], Dict[str, Any]]] = None,
|
||||
) -> None:
|
||||
self.host = host
|
||||
self.port = port
|
||||
self._server = A2AMTLSServer(cert=cert, key=key, ca=ca, host=host, port=port)
|
||||
self._executor = executor or _default_local_hermes_executor
|
||||
self._card_factory = card_factory or (lambda: _default_agent_card(self.host, self.port))
|
||||
self._tasks: Dict[str, Dict[str, Any]] = {}
|
||||
self._lock = threading.Lock()
|
||||
self._server.add_route("/.well-known/agent-card.json", self._handle_agent_card)
|
||||
self._server.add_route("/agent-card.json", self._handle_agent_card)
|
||||
self._server.add_route("/a2a/rpc", self._handle_rpc)
|
||||
|
||||
def __enter__(self) -> "A2ATaskServer":
|
||||
self.start()
|
||||
return self
|
||||
|
||||
def __exit__(self, *_: Any) -> None:
|
||||
self.stop()
|
||||
|
||||
def start(self) -> None:
|
||||
self._server.start()
|
||||
|
||||
def stop(self) -> None:
|
||||
self._server.stop()
|
||||
|
||||
def _handle_agent_card(self, payload: Dict[str, Any], *, peer_cn: str | None = None) -> Dict[str, Any]:
|
||||
return self._card_factory()
|
||||
|
||||
def _handle_rpc(self, payload: Dict[str, Any], *, peer_cn: str | None = None) -> Dict[str, Any]:
|
||||
req_id = payload.get("id")
|
||||
if payload.get("jsonrpc") != "2.0":
|
||||
return {"jsonrpc": "2.0", "id": req_id, "error": {"code": -32600, "message": "invalid jsonrpc version"}}
|
||||
|
||||
method = payload.get("method")
|
||||
params = payload.get("params") or {}
|
||||
try:
|
||||
if method == "tasks/send":
|
||||
result = self._rpc_send_task(params, peer_cn=peer_cn)
|
||||
elif method == "tasks/get":
|
||||
result = self._rpc_get_task(params)
|
||||
else:
|
||||
return {"jsonrpc": "2.0", "id": req_id, "error": {"code": -32601, "message": f"unknown method: {method}"}}
|
||||
except Exception as exc:
|
||||
logger.exception("A2A task RPC failed: %s", exc)
|
||||
return {"jsonrpc": "2.0", "id": req_id, "error": {"code": -32000, "message": str(exc)}}
|
||||
return {"jsonrpc": "2.0", "id": req_id, "result": result}
|
||||
|
||||
def _rpc_send_task(self, params: Dict[str, Any], *, peer_cn: str | None = None) -> Dict[str, Any]:
|
||||
task_text = str(params.get("task", "")).strip()
|
||||
if not task_text:
|
||||
raise ValueError("task is required")
|
||||
task_id = params.get("taskId") or uuid.uuid4().hex
|
||||
requester = params.get("requester") or peer_cn
|
||||
metadata = dict(params.get("metadata") or {})
|
||||
if peer_cn:
|
||||
metadata.setdefault("peer_cn", peer_cn)
|
||||
record = _build_task_record(task_id, task_text, requester, metadata)
|
||||
with self._lock:
|
||||
self._tasks[task_id] = record
|
||||
worker = threading.Thread(target=self._run_task, args=(task_id,), daemon=True, name=f"a2a-task-{task_id[:8]}")
|
||||
worker.start()
|
||||
return self._copy_task(task_id)
|
||||
|
||||
def _rpc_get_task(self, params: Dict[str, Any]) -> Dict[str, Any]:
|
||||
task_id = str(params.get("taskId", "")).strip()
|
||||
if not task_id:
|
||||
raise ValueError("taskId is required")
|
||||
return self._copy_task(task_id)
|
||||
|
||||
def _copy_task(self, task_id: str) -> Dict[str, Any]:
|
||||
with self._lock:
|
||||
if task_id not in self._tasks:
|
||||
raise KeyError(f"unknown taskId: {task_id}")
|
||||
return json.loads(json.dumps(self._tasks[task_id]))
|
||||
|
||||
def _run_task(self, task_id: str) -> None:
|
||||
with self._lock:
|
||||
task = self._tasks[task_id]
|
||||
task["status"] = _task_status("working", "Task is running")
|
||||
task_payload = {
|
||||
"taskId": task["taskId"],
|
||||
"task": task["task"],
|
||||
"requester": task.get("requester"),
|
||||
"metadata": dict(task.get("metadata") or {}),
|
||||
}
|
||||
try:
|
||||
result = self._executor(task_payload)
|
||||
artifact = _coerce_artifact(result)
|
||||
with self._lock:
|
||||
task = self._tasks[task_id]
|
||||
task["artifacts"] = [artifact]
|
||||
task["status"] = _task_status("completed", "Task completed")
|
||||
except Exception as exc:
|
||||
with self._lock:
|
||||
task = self._tasks[task_id]
|
||||
task["status"] = _task_status("failed", f"Task failed: {exc}")
|
||||
|
||||
|
||||
class A2ATaskClient(A2AMTLSClient):
|
||||
"""Client helper for A2A JSON-RPC task send/get flows."""
|
||||
|
||||
def discover_card(self, base_url: str) -> Dict[str, Any]:
|
||||
return self.get(f"{base_url.rstrip('/')}/.well-known/agent-card.json")
|
||||
|
||||
def _rpc_call(self, base_url: str, method: str, params: Dict[str, Any]) -> Dict[str, Any]:
|
||||
payload = {
|
||||
"jsonrpc": "2.0",
|
||||
"id": uuid.uuid4().hex,
|
||||
"method": method,
|
||||
"params": params,
|
||||
}
|
||||
response = self.post(f"{base_url.rstrip('/')}/a2a/rpc", json=payload)
|
||||
if "error" in response:
|
||||
error = response["error"]
|
||||
raise RuntimeError(error.get("message") or str(error))
|
||||
return response.get("result", {})
|
||||
|
||||
def send_task(
|
||||
self,
|
||||
base_url: str,
|
||||
*,
|
||||
task: str,
|
||||
requester: str | None = None,
|
||||
metadata: Optional[Dict[str, Any]] = None,
|
||||
) -> Dict[str, Any]:
|
||||
return self._rpc_call(
|
||||
base_url,
|
||||
"tasks/send",
|
||||
{
|
||||
"task": task,
|
||||
"requester": requester,
|
||||
"metadata": metadata or {},
|
||||
},
|
||||
)
|
||||
|
||||
def get_task(self, base_url: str, task_id: str) -> Dict[str, Any]:
|
||||
return self._rpc_call(base_url, "tasks/get", {"taskId": task_id})
|
||||
|
||||
def wait_for_task(
|
||||
self,
|
||||
base_url: str,
|
||||
task_id: str,
|
||||
*,
|
||||
timeout: float = 30.0,
|
||||
poll_interval: float = 0.5,
|
||||
) -> Dict[str, Any]:
|
||||
deadline = time.monotonic() + timeout
|
||||
while True:
|
||||
task = self.get_task(base_url, task_id)
|
||||
state = str(((task.get("status") or {}).get("state") or "")).lower()
|
||||
if state in _TERMINAL_TASK_STATES:
|
||||
return task
|
||||
if time.monotonic() >= deadline:
|
||||
raise TimeoutError(f"Timed out waiting for task {task_id}")
|
||||
time.sleep(poll_interval)
|
||||
|
||||
16
cli.py
16
cli.py
@@ -589,6 +589,7 @@ from tools.terminal_tool import set_sudo_password_callback, set_approval_callbac
|
||||
from tools.skills_tool import set_secret_capture_callback
|
||||
from hermes_cli.callbacks import prompt_for_secret
|
||||
from tools.browser_tool import _emergency_cleanup_all_sessions as _cleanup_all_browsers
|
||||
from utils import repair_and_load_json
|
||||
|
||||
# Guard to prevent cleanup from running multiple times on exit
|
||||
_cleanup_done = False
|
||||
@@ -3569,7 +3570,11 @@ class HermesCLI:
|
||||
result_json = _asyncio.run(
|
||||
vision_analyze_tool(image_url=str(img_path), user_prompt=analysis_prompt)
|
||||
)
|
||||
result = _json.loads(result_json)
|
||||
result = repair_and_load_json(
|
||||
result_json,
|
||||
default={},
|
||||
context="cli_image_analysis",
|
||||
) if isinstance(result_json, str) else {}
|
||||
if result.get("success"):
|
||||
description = result.get("analysis", "")
|
||||
enriched_parts.append(
|
||||
@@ -4960,7 +4965,14 @@ class HermesCLI:
|
||||
from tools.cronjob_tools import cronjob as cronjob_tool
|
||||
|
||||
def _cron_api(**kwargs):
|
||||
return json.loads(cronjob_tool(**kwargs))
|
||||
result = repair_and_load_json(
|
||||
cronjob_tool(**kwargs),
|
||||
default=None,
|
||||
context="cli_cron_command",
|
||||
)
|
||||
if isinstance(result, dict):
|
||||
return result
|
||||
return {"success": False, "error": "Invalid JSON from cronjob tool"}
|
||||
|
||||
def _normalize_skills(values):
|
||||
normalized = []
|
||||
|
||||
@@ -1,132 +0,0 @@
|
||||
"""CLI helpers for A2A task delegation."""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import json
|
||||
import os
|
||||
import re
|
||||
import sys
|
||||
import time
|
||||
from pathlib import Path
|
||||
from typing import Any
|
||||
|
||||
from agent.a2a_mtls import A2ATaskClient, A2ATaskServer
|
||||
from hermes_cli.config import get_hermes_home
|
||||
|
||||
|
||||
def _registry_path() -> Path:
|
||||
return get_hermes_home() / "a2a_agents.json"
|
||||
|
||||
|
||||
def _default_identity_paths() -> tuple[str, str, str]:
|
||||
hermes_home = get_hermes_home()
|
||||
agent_name = os.environ.get("HERMES_AGENT_NAME", "hermes").lower()
|
||||
cert = os.environ.get(
|
||||
"HERMES_A2A_CERT",
|
||||
str(hermes_home / "pki" / "agents" / agent_name / f"{agent_name}.crt"),
|
||||
)
|
||||
key = os.environ.get(
|
||||
"HERMES_A2A_KEY",
|
||||
str(hermes_home / "pki" / "agents" / agent_name / f"{agent_name}.key"),
|
||||
)
|
||||
ca = os.environ.get(
|
||||
"HERMES_A2A_CA",
|
||||
str(hermes_home / "pki" / "ca" / "fleet-ca.crt"),
|
||||
)
|
||||
return cert, key, ca
|
||||
|
||||
|
||||
def load_agent_registry(path: Path | None = None) -> dict[str, Any]:
|
||||
registry_path = path or _registry_path()
|
||||
if not registry_path.exists():
|
||||
return {}
|
||||
return json.loads(registry_path.read_text(encoding="utf-8"))
|
||||
|
||||
|
||||
def resolve_agent_url(agent: str, *, registry_path: Path | None = None) -> str:
|
||||
key = re.sub(r"[^A-Za-z0-9]+", "_", agent).upper()
|
||||
env_value = os.getenv(f"HERMES_A2A_{key}_URL")
|
||||
if env_value:
|
||||
return env_value
|
||||
|
||||
registry = load_agent_registry(registry_path)
|
||||
entry = registry.get(agent)
|
||||
if isinstance(entry, str) and entry:
|
||||
return entry
|
||||
if isinstance(entry, dict):
|
||||
url = entry.get("url") or entry.get("base_url") or entry.get("card_url")
|
||||
if url:
|
||||
return str(url)
|
||||
if agent.startswith("https://") or agent.startswith("http://"):
|
||||
return agent
|
||||
raise SystemExit(f"Unknown A2A agent '{agent}'. Set HERMES_A2A_{key}_URL or add it to {_registry_path()}.")
|
||||
|
||||
|
||||
def _print(data: dict[str, Any]) -> None:
|
||||
print(json.dumps(data, indent=2, ensure_ascii=False))
|
||||
|
||||
|
||||
def cmd_send(args) -> None:
|
||||
base_url = args.url or resolve_agent_url(args.agent)
|
||||
cert, key, ca = args.cert, args.key, args.ca
|
||||
if not (cert and key and ca):
|
||||
cert, key, ca = _default_identity_paths()
|
||||
client = A2ATaskClient(cert=cert, key=key, ca=ca)
|
||||
card = client.discover_card(base_url)
|
||||
task = client.send_task(
|
||||
base_url,
|
||||
task=args.task,
|
||||
requester=args.requester,
|
||||
metadata={"agent": args.agent},
|
||||
)
|
||||
if args.wait:
|
||||
task = client.wait_for_task(
|
||||
base_url,
|
||||
task["taskId"],
|
||||
timeout=args.timeout,
|
||||
poll_interval=args.poll_interval,
|
||||
)
|
||||
_print({
|
||||
"agent": args.agent,
|
||||
"url": base_url,
|
||||
"card": card,
|
||||
"task": task,
|
||||
})
|
||||
|
||||
|
||||
def cmd_status(args) -> None:
|
||||
base_url = args.url or resolve_agent_url(args.agent)
|
||||
cert, key, ca = args.cert, args.key, args.ca
|
||||
if not (cert and key and ca):
|
||||
cert, key, ca = _default_identity_paths()
|
||||
client = A2ATaskClient(cert=cert, key=key, ca=ca)
|
||||
task = client.get_task(base_url, args.task_id)
|
||||
_print({"agent": args.agent, "url": base_url, "task": task})
|
||||
|
||||
|
||||
def cmd_serve(args) -> None:
|
||||
cert, key, ca = args.cert, args.key, args.ca
|
||||
if not (cert and key and ca):
|
||||
cert, key, ca = _default_identity_paths()
|
||||
server = A2ATaskServer(cert=cert, key=key, ca=ca, host=args.host, port=args.port)
|
||||
server.start()
|
||||
print(f"A2A task server listening on https://{args.host}:{args.port}")
|
||||
try:
|
||||
while True:
|
||||
time.sleep(1)
|
||||
except KeyboardInterrupt:
|
||||
server.stop()
|
||||
|
||||
|
||||
def cmd_a2a(args) -> None:
|
||||
command = getattr(args, "a2a_command", None) or "send"
|
||||
if command == "send":
|
||||
cmd_send(args)
|
||||
return
|
||||
if command == "status":
|
||||
cmd_status(args)
|
||||
return
|
||||
if command == "serve":
|
||||
cmd_serve(args)
|
||||
return
|
||||
raise SystemExit(f"Unknown a2a command: {command}")
|
||||
@@ -173,13 +173,6 @@ from hermes_constants import OPENROUTER_BASE_URL
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
|
||||
def cmd_a2a(args):
|
||||
"""Dispatch A2A CLI subcommands lazily to avoid heavy imports at startup."""
|
||||
from hermes_cli.a2a_cmd import cmd_a2a as _cmd_a2a
|
||||
|
||||
return _cmd_a2a(args)
|
||||
|
||||
|
||||
def _relative_time(ts) -> str:
|
||||
"""Format a timestamp as relative time (e.g., '2h ago', 'yesterday')."""
|
||||
if not ts:
|
||||
@@ -4788,45 +4781,6 @@ For more help on a command:
|
||||
|
||||
gateway_parser.set_defaults(func=cmd_gateway)
|
||||
|
||||
# =========================================================================
|
||||
# a2a command
|
||||
# =========================================================================
|
||||
a2a_parser = subparsers.add_parser(
|
||||
"a2a",
|
||||
help="A2A task delegation over mutual TLS",
|
||||
description="Send, inspect, and serve structured A2A tasks between Hermes agents",
|
||||
)
|
||||
a2a_subparsers = a2a_parser.add_subparsers(dest="a2a_command")
|
||||
|
||||
a2a_send = a2a_subparsers.add_parser("send", help="Send an A2A task to another agent")
|
||||
a2a_send.add_argument("--agent", required=True, help="Agent alias or URL (for example: allegro)")
|
||||
a2a_send.add_argument("--task", required=True, help="Task text to delegate")
|
||||
a2a_send.add_argument("--url", help="Explicit base URL for the remote agent")
|
||||
a2a_send.add_argument("--requester", default=None, help="Requester label included in task metadata")
|
||||
a2a_send.add_argument("--wait", action="store_true", help="Poll until the task reaches a terminal state")
|
||||
a2a_send.add_argument("--timeout", type=float, default=30.0, help="Wait timeout in seconds (default: 30)")
|
||||
a2a_send.add_argument("--poll-interval", type=float, default=0.5, help="Polling interval in seconds while waiting (default: 0.5)")
|
||||
a2a_send.add_argument("--cert", default=None, help="Client certificate path (defaults from HERMES_A2A_CERT)")
|
||||
a2a_send.add_argument("--key", default=None, help="Client private key path (defaults from HERMES_A2A_KEY)")
|
||||
a2a_send.add_argument("--ca", default=None, help="Fleet CA certificate path (defaults from HERMES_A2A_CA)")
|
||||
|
||||
a2a_status = a2a_subparsers.add_parser("status", help="Fetch the current status of an A2A task")
|
||||
a2a_status.add_argument("--agent", required=True, help="Agent alias or URL (for example: allegro)")
|
||||
a2a_status.add_argument("--task-id", required=True, help="Task identifier returned by a2a send")
|
||||
a2a_status.add_argument("--url", help="Explicit base URL for the remote agent")
|
||||
a2a_status.add_argument("--cert", default=None, help="Client certificate path (defaults from HERMES_A2A_CERT)")
|
||||
a2a_status.add_argument("--key", default=None, help="Client private key path (defaults from HERMES_A2A_KEY)")
|
||||
a2a_status.add_argument("--ca", default=None, help="Fleet CA certificate path (defaults from HERMES_A2A_CA)")
|
||||
|
||||
a2a_serve = a2a_subparsers.add_parser("serve", help="Run the local A2A task server")
|
||||
a2a_serve.add_argument("--host", default=os.environ.get("HERMES_A2A_HOST", "127.0.0.1"), help="Bind host (default: HERMES_A2A_HOST or 127.0.0.1)")
|
||||
a2a_serve.add_argument("--port", type=int, default=int(os.environ.get("HERMES_A2A_PORT", "9443")), help="Bind port (default: HERMES_A2A_PORT or 9443)")
|
||||
a2a_serve.add_argument("--cert", default=None, help="Server certificate path (defaults from HERMES_A2A_CERT)")
|
||||
a2a_serve.add_argument("--key", default=None, help="Server private key path (defaults from HERMES_A2A_KEY)")
|
||||
a2a_serve.add_argument("--ca", default=None, help="Fleet CA certificate path (defaults from HERMES_A2A_CA)")
|
||||
|
||||
a2a_parser.set_defaults(func=cmd_a2a)
|
||||
|
||||
# =========================================================================
|
||||
# setup command
|
||||
# =========================================================================
|
||||
|
||||
@@ -572,94 +572,3 @@ class TestA2AMTLSServerAndClient:
|
||||
|
||||
assert not errors, f"Concurrent connection errors: {errors}"
|
||||
assert len(results) == 3
|
||||
|
||||
|
||||
@_requires_crypto
|
||||
class TestA2ATaskServerAndClient:
|
||||
"""Structured A2A task send/get flow over mTLS."""
|
||||
|
||||
@pytest.fixture(autouse=True)
|
||||
def _pki(self, tmp_path):
|
||||
ca_dir = tmp_path / "ca"
|
||||
ca_dir.mkdir()
|
||||
self.ca_crt, self.ca_key = _make_ca_keypair(ca_dir)
|
||||
agent_dir = tmp_path / "agents"
|
||||
agent_dir.mkdir()
|
||||
self.srv_crt, self.srv_key = _make_agent_keypair(
|
||||
agent_dir, "timmy", self.ca_crt, self.ca_key
|
||||
)
|
||||
self.cli_crt, self.cli_key = _make_agent_keypair(
|
||||
agent_dir, "allegro", self.ca_crt, self.ca_key
|
||||
)
|
||||
|
||||
@pytest.fixture()
|
||||
def task_server(self):
|
||||
from agent.a2a_mtls import A2ATaskServer
|
||||
|
||||
gate = threading.Event()
|
||||
|
||||
def analyze_executor(task: dict[str, object]) -> dict[str, object]:
|
||||
gate.wait(timeout=2)
|
||||
text = str(task.get("task", ""))
|
||||
return {
|
||||
"text": f"analysis:{text}",
|
||||
"metadata": {"tool": "local-hermes-stub"},
|
||||
}
|
||||
|
||||
port = _find_free_port()
|
||||
server = A2ATaskServer(
|
||||
cert=self.srv_crt,
|
||||
key=self.srv_key,
|
||||
ca=self.ca_crt,
|
||||
host="127.0.0.1",
|
||||
port=port,
|
||||
executor=analyze_executor,
|
||||
)
|
||||
with server:
|
||||
time.sleep(0.1)
|
||||
yield server, port, gate
|
||||
|
||||
def test_task_send_get_and_completion_flow(self, task_server):
|
||||
from agent.a2a_mtls import A2ATaskClient
|
||||
|
||||
server, port, gate = task_server
|
||||
client = A2ATaskClient(cert=self.cli_crt, key=self.cli_key, ca=self.ca_crt)
|
||||
base_url = f"https://127.0.0.1:{port}"
|
||||
|
||||
card = client.discover_card(base_url)
|
||||
assert card["name"]
|
||||
|
||||
submitted = client.send_task(base_url, task="Analyze README.md", requester="timmy")
|
||||
assert submitted["status"]["state"] in {"submitted", "working"}
|
||||
|
||||
in_flight = client.get_task(base_url, submitted["taskId"])
|
||||
assert in_flight["status"]["state"] in {"submitted", "working"}
|
||||
|
||||
gate.set()
|
||||
completed = client.wait_for_task(base_url, submitted["taskId"], timeout=5.0, poll_interval=0.05)
|
||||
assert completed["status"]["state"] == "completed"
|
||||
assert completed["artifacts"][0]["text"] == "analysis:Analyze README.md"
|
||||
|
||||
def test_failed_executor_marks_task_failed(self):
|
||||
from agent.a2a_mtls import A2ATaskClient, A2ATaskServer
|
||||
|
||||
def failing_executor(task: dict[str, object]) -> dict[str, object]:
|
||||
raise RuntimeError("boom")
|
||||
|
||||
port = _find_free_port()
|
||||
server = A2ATaskServer(
|
||||
cert=self.srv_crt,
|
||||
key=self.srv_key,
|
||||
ca=self.ca_crt,
|
||||
host="127.0.0.1",
|
||||
port=port,
|
||||
executor=failing_executor,
|
||||
)
|
||||
with server:
|
||||
time.sleep(0.1)
|
||||
client = A2ATaskClient(cert=self.cli_crt, key=self.cli_key, ca=self.ca_crt)
|
||||
base_url = f"https://127.0.0.1:{port}"
|
||||
submitted = client.send_task(base_url, task="explode", requester="timmy")
|
||||
failed = client.wait_for_task(base_url, submitted["taskId"], timeout=5.0, poll_interval=0.05)
|
||||
assert failed["status"]["state"] == "failed"
|
||||
assert "boom" in failed["status"]["message"]
|
||||
|
||||
62
tests/cli/test_cli_json_repair.py
Normal file
62
tests/cli/test_cli_json_repair.py
Normal file
@@ -0,0 +1,62 @@
|
||||
import sys
|
||||
import types
|
||||
from unittest.mock import patch
|
||||
|
||||
|
||||
def _stub_auxiliary_client():
|
||||
stub = types.ModuleType("agent.auxiliary_client")
|
||||
stub.call_llm = lambda *args, **kwargs: None
|
||||
stub.resolve_provider_client = lambda *args, **kwargs: (None, None)
|
||||
stub.get_text_auxiliary_client = lambda *args, **kwargs: (None, None)
|
||||
stub.async_call_llm = lambda *args, **kwargs: None
|
||||
stub.extract_content_or_reasoning = lambda *args, **kwargs: ""
|
||||
stub._OR_HEADERS = {}
|
||||
stub._get_task_timeout = lambda *args, **kwargs: 30
|
||||
sys.modules["agent.auxiliary_client"] = stub
|
||||
|
||||
|
||||
def _stub_vision_tools(vision_analyze_tool):
|
||||
stub = types.ModuleType("tools.vision_tools")
|
||||
stub.vision_analyze_tool = vision_analyze_tool
|
||||
sys.modules["tools.vision_tools"] = stub
|
||||
|
||||
|
||||
def test_preprocess_images_with_vision_repairs_malformed_json(tmp_path):
|
||||
_stub_auxiliary_client()
|
||||
from cli import HermesCLI
|
||||
|
||||
cli_obj = HermesCLI.__new__(HermesCLI)
|
||||
image_path = tmp_path / "test.png"
|
||||
image_path.write_bytes(b"fake-image-bytes")
|
||||
|
||||
async def fake_vision(**kwargs):
|
||||
return "{'success': true, 'analysis': 'Recovered image description',}"
|
||||
|
||||
_stub_vision_tools(fake_vision)
|
||||
result = HermesCLI._preprocess_images_with_vision(
|
||||
cli_obj,
|
||||
"Describe this",
|
||||
[image_path],
|
||||
announce=False,
|
||||
)
|
||||
|
||||
assert "Recovered image description" in result
|
||||
assert "Describe this" in result
|
||||
assert str(image_path) in result
|
||||
|
||||
|
||||
def test_handle_cron_command_repairs_malformed_json(capsys):
|
||||
_stub_auxiliary_client()
|
||||
from cli import HermesCLI
|
||||
|
||||
cli_obj = HermesCLI.__new__(HermesCLI)
|
||||
malformed_result = """{'success': true, 'jobs': [{'job_id': 'job-1234567890ab', 'name': 'Nightly Check', 'state': 'scheduled', 'schedule': 'every 1h', 'repeat': 'forever', 'prompt_preview': 'Check server status', 'skills': ['blogwatcher',], 'next_run_at': '2026-04-22T01:00:00Z',},],}"""
|
||||
|
||||
with patch("tools.cronjob_tools.cronjob", return_value=malformed_result):
|
||||
HermesCLI._handle_cron_command(cli_obj, "/cron list")
|
||||
|
||||
out = capsys.readouterr().out
|
||||
assert "Scheduled Jobs:" in out
|
||||
assert "job-1234567890ab" in out
|
||||
assert "Nightly Check" in out
|
||||
assert "blogwatcher" in out
|
||||
@@ -1,95 +0,0 @@
|
||||
from __future__ import annotations
|
||||
|
||||
import argparse
|
||||
import json
|
||||
from pathlib import Path
|
||||
from unittest.mock import patch
|
||||
|
||||
import pytest
|
||||
|
||||
|
||||
def test_cmd_send_uses_registry_and_waits_for_terminal_task(tmp_path, monkeypatch, capsys):
|
||||
hermes_home = tmp_path / ".hermes"
|
||||
hermes_home.mkdir()
|
||||
(hermes_home / "a2a_agents.json").write_text(
|
||||
json.dumps({"allegro": {"url": "https://127.0.0.1:9443"}}),
|
||||
encoding="utf-8",
|
||||
)
|
||||
monkeypatch.setenv("HERMES_HOME", str(hermes_home))
|
||||
|
||||
from hermes_cli.a2a_cmd import cmd_a2a
|
||||
|
||||
class FakeClient:
|
||||
def __init__(self, **kwargs):
|
||||
self.kwargs = kwargs
|
||||
|
||||
def discover_card(self, base_url: str):
|
||||
assert base_url == "https://127.0.0.1:9443"
|
||||
return {"name": "allegro", "url": base_url}
|
||||
|
||||
def send_task(self, base_url: str, *, task: str, requester: str | None = None, metadata=None):
|
||||
assert task == "analyze README"
|
||||
return {"taskId": "task-123", "status": {"state": "submitted"}}
|
||||
|
||||
def wait_for_task(self, base_url: str, task_id: str, *, timeout: float, poll_interval: float):
|
||||
assert task_id == "task-123"
|
||||
return {
|
||||
"taskId": task_id,
|
||||
"status": {"state": "completed"},
|
||||
"artifacts": [{"text": "README looks healthy"}],
|
||||
}
|
||||
|
||||
args = argparse.Namespace(
|
||||
a2a_command="send",
|
||||
agent="allegro",
|
||||
task="analyze README",
|
||||
url=None,
|
||||
wait=True,
|
||||
timeout=5.0,
|
||||
poll_interval=0.01,
|
||||
requester="timmy",
|
||||
cert="cert.pem",
|
||||
key="key.pem",
|
||||
ca="ca.pem",
|
||||
)
|
||||
|
||||
with patch("hermes_cli.a2a_cmd.A2ATaskClient", FakeClient):
|
||||
cmd_a2a(args)
|
||||
|
||||
result = json.loads(capsys.readouterr().out)
|
||||
assert result["agent"] == "allegro"
|
||||
assert result["card"]["name"] == "allegro"
|
||||
assert result["task"]["status"]["state"] == "completed"
|
||||
assert result["task"]["artifacts"][0]["text"] == "README looks healthy"
|
||||
|
||||
|
||||
def test_resolve_agent_url_supports_env_override(monkeypatch):
|
||||
monkeypatch.setenv("HERMES_A2A_ALLEGRO_URL", "https://fleet-allegro:9443")
|
||||
from hermes_cli.a2a_cmd import resolve_agent_url
|
||||
|
||||
assert resolve_agent_url("allegro") == "https://fleet-allegro:9443"
|
||||
|
||||
|
||||
def test_cmd_send_requires_known_agent(tmp_path, monkeypatch):
|
||||
hermes_home = tmp_path / ".hermes"
|
||||
hermes_home.mkdir()
|
||||
monkeypatch.setenv("HERMES_HOME", str(hermes_home))
|
||||
|
||||
from hermes_cli.a2a_cmd import cmd_a2a
|
||||
|
||||
args = argparse.Namespace(
|
||||
a2a_command="send",
|
||||
agent="unknown",
|
||||
task="do work",
|
||||
url=None,
|
||||
wait=False,
|
||||
timeout=5.0,
|
||||
poll_interval=0.05,
|
||||
requester=None,
|
||||
cert="cert.pem",
|
||||
key="key.pem",
|
||||
ca="ca.pem",
|
||||
)
|
||||
|
||||
with pytest.raises(SystemExit):
|
||||
cmd_a2a(args)
|
||||
108
tests/tools/test_browser_json_repair.py
Normal file
108
tests/tools/test_browser_json_repair.py
Normal file
@@ -0,0 +1,108 @@
|
||||
import io
|
||||
import json
|
||||
import sys
|
||||
import types
|
||||
from unittest.mock import MagicMock, patch
|
||||
|
||||
|
||||
def _stub_auxiliary_client():
|
||||
stub = types.ModuleType("agent.auxiliary_client")
|
||||
stub.call_llm = lambda *args, **kwargs: None
|
||||
stub.resolve_provider_client = lambda *args, **kwargs: (None, None)
|
||||
stub.get_text_auxiliary_client = lambda *args, **kwargs: (None, None)
|
||||
stub.async_call_llm = lambda *args, **kwargs: None
|
||||
stub.extract_content_or_reasoning = lambda *args, **kwargs: ""
|
||||
stub._OR_HEADERS = {}
|
||||
stub._get_task_timeout = lambda *args, **kwargs: 30
|
||||
sys.modules["agent.auxiliary_client"] = stub
|
||||
|
||||
|
||||
def test_run_browser_command_repairs_malformed_stdout_envelope(tmp_path):
|
||||
_stub_auxiliary_client()
|
||||
from tools.browser_tool import _run_browser_command
|
||||
|
||||
mock_proc = MagicMock()
|
||||
mock_proc.returncode = 0
|
||||
mock_proc.wait.return_value = 0
|
||||
fake_session = {
|
||||
"session_name": "test-session",
|
||||
"session_id": "test-id",
|
||||
"cdp_url": None,
|
||||
}
|
||||
malformed_stdout = "{'success': true, 'data': {'url': 'https://example.com',},}"
|
||||
|
||||
def fake_open(path, mode="r", *args, **kwargs):
|
||||
path = str(path)
|
||||
if path.endswith("_stdout_navigate"):
|
||||
return io.StringIO(malformed_stdout)
|
||||
if path.endswith("_stderr_navigate"):
|
||||
return io.StringIO("")
|
||||
raise FileNotFoundError(path)
|
||||
|
||||
with (
|
||||
patch("tools.browser_tool._find_agent_browser", return_value="/usr/bin/agent-browser"),
|
||||
patch("tools.browser_tool._get_session_info", return_value=fake_session),
|
||||
patch("tools.browser_tool._socket_safe_tmpdir", return_value=str(tmp_path)),
|
||||
patch("tools.browser_tool._merge_browser_path", side_effect=lambda p: p),
|
||||
patch("tools.interrupt.is_interrupted", return_value=False),
|
||||
patch("subprocess.Popen", return_value=mock_proc),
|
||||
patch("os.open", return_value=99),
|
||||
patch("os.close"),
|
||||
patch("os.unlink"),
|
||||
patch("builtins.open", side_effect=fake_open),
|
||||
):
|
||||
result = _run_browser_command("task-1", "navigate", ["https://example.com"])
|
||||
|
||||
assert result["success"] is True
|
||||
assert result["data"]["url"] == "https://example.com"
|
||||
|
||||
|
||||
def test_agent_browser_eval_repairs_malformed_json_result():
|
||||
_stub_auxiliary_client()
|
||||
from tools.browser_tool import _browser_eval
|
||||
|
||||
with patch(
|
||||
"tools.browser_tool._run_browser_command",
|
||||
return_value={"success": True, "data": {"result": "{'items': ['a', 'b',],}"}},
|
||||
):
|
||||
result = json.loads(_browser_eval("document.body.innerText", task_id="test"))
|
||||
|
||||
assert result["success"] is True
|
||||
assert result["result"] == {"items": ["a", "b"]}
|
||||
assert result["result_type"] == "dict"
|
||||
|
||||
|
||||
def test_camofox_eval_repairs_malformed_json_result():
|
||||
_stub_auxiliary_client()
|
||||
from tools.browser_tool import _camofox_eval
|
||||
|
||||
with (
|
||||
patch("tools.browser_camofox._ensure_tab", return_value={"tab_id": "tab-1", "user_id": "user-1"}),
|
||||
patch("tools.browser_camofox._post", return_value={"result": "{'count': 3,}"}),
|
||||
):
|
||||
result = json.loads(_camofox_eval("2+1", task_id="test"))
|
||||
|
||||
assert result["success"] is True
|
||||
assert result["result"] == {"count": 3}
|
||||
assert result["result_type"] == "dict"
|
||||
|
||||
|
||||
def test_browser_get_images_repairs_malformed_json_result():
|
||||
_stub_auxiliary_client()
|
||||
from tools.browser_tool import browser_get_images
|
||||
|
||||
with patch(
|
||||
"tools.browser_tool._run_browser_command",
|
||||
return_value={
|
||||
"success": True,
|
||||
"data": {
|
||||
"result": "[{\"src\": \"https://example.com/cat.png\", \"alt\": \"cat\",}]"
|
||||
},
|
||||
},
|
||||
):
|
||||
result = json.loads(browser_get_images(task_id="test"))
|
||||
|
||||
assert result["success"] is True
|
||||
assert result["count"] == 1
|
||||
assert result["images"] == [{"src": "https://example.com/cat.png", "alt": "cat"}]
|
||||
assert "warning" not in result
|
||||
@@ -67,6 +67,7 @@ from typing import Dict, Any, Optional, List
|
||||
from pathlib import Path
|
||||
from agent.auxiliary_client import call_llm
|
||||
from hermes_constants import get_hermes_home
|
||||
from utils import repair_and_load_json
|
||||
|
||||
try:
|
||||
from tools.website_policy import check_website_access
|
||||
@@ -1171,8 +1172,12 @@ def _run_browser_command(
|
||||
return {"success": False, "error": f"Browser command '{command}' returned no output"}
|
||||
|
||||
if stdout_text:
|
||||
try:
|
||||
parsed = json.loads(stdout_text)
|
||||
parsed = repair_and_load_json(
|
||||
stdout_text,
|
||||
default=None,
|
||||
context=f"browser_{command}_stdout",
|
||||
)
|
||||
if isinstance(parsed, dict):
|
||||
# Warn if snapshot came back empty (common sign of daemon/CDP issues)
|
||||
if command == "snapshot" and parsed.get("success"):
|
||||
snap_data = parsed.get("data", {})
|
||||
@@ -1181,35 +1186,35 @@ def _run_browser_command(
|
||||
"Possible stale daemon or CDP connection issue. "
|
||||
"returncode=%s", returncode)
|
||||
return parsed
|
||||
except json.JSONDecodeError:
|
||||
raw = stdout_text[:2000]
|
||||
logger.warning("browser '%s' returned non-JSON output (rc=%s): %s",
|
||||
command, returncode, raw[:500])
|
||||
|
||||
if command == "screenshot":
|
||||
stderr_text = (stderr or "").strip()
|
||||
combined_text = "\n".join(
|
||||
part for part in [stdout_text, stderr_text] if part
|
||||
raw = stdout_text[:2000]
|
||||
logger.warning("browser '%s' returned non-JSON output (rc=%s): %s",
|
||||
command, returncode, raw[:500])
|
||||
|
||||
if command == "screenshot":
|
||||
stderr_text = (stderr or "").strip()
|
||||
combined_text = "\n".join(
|
||||
part for part in [stdout_text, stderr_text] if part
|
||||
)
|
||||
recovered_path = _extract_screenshot_path_from_text(combined_text)
|
||||
|
||||
if recovered_path and Path(recovered_path).exists():
|
||||
logger.info(
|
||||
"browser 'screenshot' recovered file from non-JSON output: %s",
|
||||
recovered_path,
|
||||
)
|
||||
recovered_path = _extract_screenshot_path_from_text(combined_text)
|
||||
return {
|
||||
"success": True,
|
||||
"data": {
|
||||
"path": recovered_path,
|
||||
"raw": raw,
|
||||
},
|
||||
}
|
||||
|
||||
if recovered_path and Path(recovered_path).exists():
|
||||
logger.info(
|
||||
"browser 'screenshot' recovered file from non-JSON output: %s",
|
||||
recovered_path,
|
||||
)
|
||||
return {
|
||||
"success": True,
|
||||
"data": {
|
||||
"path": recovered_path,
|
||||
"raw": raw,
|
||||
},
|
||||
}
|
||||
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Non-JSON output from agent-browser for '{command}': {raw}"
|
||||
}
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Non-JSON output from agent-browser for '{command}': {raw}"
|
||||
}
|
||||
|
||||
# Check for errors
|
||||
if returncode != 0:
|
||||
@@ -1777,10 +1782,11 @@ def _browser_eval(expression: str, task_id: Optional[str] = None) -> str:
|
||||
# is valid JSON, parse it so the model gets structured data.
|
||||
parsed = raw_result
|
||||
if isinstance(raw_result, str):
|
||||
try:
|
||||
parsed = json.loads(raw_result)
|
||||
except (json.JSONDecodeError, ValueError):
|
||||
pass # keep as string
|
||||
parsed = repair_and_load_json(
|
||||
raw_result,
|
||||
default=raw_result,
|
||||
context="browser_eval_result",
|
||||
)
|
||||
|
||||
return json.dumps({
|
||||
"success": True,
|
||||
@@ -1801,10 +1807,11 @@ def _camofox_eval(expression: str, task_id: Optional[str] = None) -> str:
|
||||
raw_result = resp.get("result") if isinstance(resp, dict) else resp
|
||||
parsed = raw_result
|
||||
if isinstance(raw_result, str):
|
||||
try:
|
||||
parsed = json.loads(raw_result)
|
||||
except (json.JSONDecodeError, ValueError):
|
||||
pass
|
||||
parsed = repair_and_load_json(
|
||||
raw_result,
|
||||
default=raw_result,
|
||||
context="camofox_eval_result",
|
||||
)
|
||||
|
||||
return json.dumps({
|
||||
"success": True,
|
||||
@@ -1904,26 +1911,29 @@ def browser_get_images(task_id: Optional[str] = None) -> str:
|
||||
if result.get("success"):
|
||||
data = result.get("data", {})
|
||||
raw_result = data.get("result", "[]")
|
||||
|
||||
try:
|
||||
# Parse the JSON string returned by JavaScript
|
||||
if isinstance(raw_result, str):
|
||||
images = json.loads(raw_result)
|
||||
else:
|
||||
images = raw_result
|
||||
|
||||
return json.dumps({
|
||||
"success": True,
|
||||
"images": images,
|
||||
"count": len(images)
|
||||
}, ensure_ascii=False)
|
||||
except json.JSONDecodeError:
|
||||
return json.dumps({
|
||||
"success": True,
|
||||
"images": [],
|
||||
"count": 0,
|
||||
"warning": "Could not parse image data"
|
||||
}, ensure_ascii=False)
|
||||
|
||||
warning = None
|
||||
if isinstance(raw_result, str):
|
||||
images = repair_and_load_json(
|
||||
raw_result,
|
||||
default=None,
|
||||
context="browser_get_images_result",
|
||||
)
|
||||
else:
|
||||
images = raw_result
|
||||
|
||||
if not isinstance(images, list):
|
||||
images = []
|
||||
warning = "Could not parse image data"
|
||||
|
||||
payload = {
|
||||
"success": True,
|
||||
"images": images,
|
||||
"count": len(images),
|
||||
}
|
||||
if warning:
|
||||
payload["warning"] = warning
|
||||
return json.dumps(payload, ensure_ascii=False)
|
||||
else:
|
||||
return json.dumps({
|
||||
"success": False,
|
||||
|
||||
Reference in New Issue
Block a user