ling/hermes-agent
1
0
Fork 0
mirror of https://github.com/NousResearch/hermes-agent.git synced 2026-04-28 06:51:16 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix: return JSON parse error to model instead of dispatching with empty args (#2342)

Browse Source 
When the model produces malformed JSON in tool call arguments, the agent
loop was setting args={} and dispatching the tool anyway, wasting an
iteration and producing a confusing downstream error. Now the error is
returned directly as the tool result so the model can retry with valid JSON.

Co-authored-by: alireza78a <alireza78.crypto@gmail.com>
This commit is contained in:
Teknium
2026-03-21 09:56:44 -07:00
committed by GitHub
parent 3835a8d5df
commit aefcdd6f7f
1 changed files with 72 additions and 61 deletions
Download Patch File Download Diff File Expand all files Collapse all files

133
environments/agent_loop.py
View File

@@ -346,78 +346,89 @@ class HermesAgentLoop:
tool_name, turn + 1, tool_name, turn + 1,
) )
else: else:
# Parse arguments and dispatch # Parse arguments
try: try:
args = json.loads(tool_args_raw) args = json.loads(tool_args_raw)
except json.JSONDecodeError: except json.JSONDecodeError as e:
args = {} args = None
tool_result = json.dumps(
{"error": f"Invalid JSON in tool arguments: {e}. Please retry with valid JSON."}
)
tool_errors.append(ToolError(
turn=turn + 1, tool_name=tool_name,
arguments=tool_args_raw[:200],
error=f"Invalid JSON: {e}",
tool_result=tool_result,
))
logger.warning( logger.warning(
"Invalid JSON in tool call arguments for '%s': %s", "Invalid JSON in tool call arguments for '%s': %s",
tool_name, tool_args_raw[:200], tool_name, tool_args_raw[:200],
) )
try: # Dispatch tool only if arguments parsed successfully
if tool_name == "terminal": if args is not None:
backend = os.getenv("TERMINAL_ENV", "local") try:
cmd_preview = args.get("command", "")[:80] if tool_name == "terminal":
logger.info( backend = os.getenv("TERMINAL_ENV", "local")
"[%s] $ %s", self.task_id[:8], cmd_preview, cmd_preview = args.get("command", "")[:80]
) logger.info(
"[%s] $ %s", self.task_id[:8], cmd_preview,
)
tool_submit_time = _time.monotonic() tool_submit_time = _time.monotonic()
# Todo tool -- handle locally (needs per-loop TodoStore) # Todo tool -- handle locally (needs per-loop TodoStore)
if tool_name == "todo": if tool_name == "todo":
tool_result = _todo_tool( tool_result = _todo_tool(
todos=args.get("todos"), todos=args.get("todos"),
merge=args.get("merge", False), merge=args.get("merge", False),
store=_todo_store, store=_todo_store,
) )
tool_elapsed = _time.monotonic() - tool_submit_time tool_elapsed = _time.monotonic() - tool_submit_time
elif tool_name == "memory": elif tool_name == "memory":
tool_result = json.dumps({"error": "Memory is not available in RL environments."}) tool_result = json.dumps({"error": "Memory is not available in RL environments."})
tool_elapsed = _time.monotonic() - tool_submit_time tool_elapsed = _time.monotonic() - tool_submit_time
elif tool_name == "session_search": elif tool_name == "session_search":
tool_result = json.dumps({"error": "Session search is not available in RL environments."}) tool_result = json.dumps({"error": "Session search is not available in RL environments."})
tool_elapsed = _time.monotonic() - tool_submit_time tool_elapsed = _time.monotonic() - tool_submit_time
else: else:
# Run tool calls in a thread pool so backends that # Run tool calls in a thread pool so backends that
# use asyncio.run() internally (modal, docker, daytona) get # use asyncio.run() internally (modal, docker, daytona) get
# a clean event loop instead of deadlocking. # a clean event loop instead of deadlocking.
loop = asyncio.get_event_loop() loop = asyncio.get_event_loop()
# Capture current tool_name/args for the lambda # Capture current tool_name/args for the lambda
_tn, _ta, _tid = tool_name, args, self.task_id _tn, _ta, _tid = tool_name, args, self.task_id
tool_result = await loop.run_in_executor( tool_result = await loop.run_in_executor(
_tool_executor, _tool_executor,
lambda: handle_function_call( lambda: handle_function_call(
_tn, _ta, task_id=_tid, _tn, _ta, task_id=_tid,
user_task=_user_task, user_task=_user_task,
), ),
) )
tool_elapsed = _time.monotonic() - tool_submit_time tool_elapsed = _time.monotonic() - tool_submit_time
# Log slow tools and thread pool stats for debugging # Log slow tools and thread pool stats for debugging
pool_active = _tool_executor._work_queue.qsize() pool_active = _tool_executor._work_queue.qsize()
if tool_elapsed > 30: if tool_elapsed > 30:
logger.warning( logger.warning(
"[%s] turn %d: %s took %.1fs (pool queue=%d)", "[%s] turn %d: %s took %.1fs (pool queue=%d)",
self.task_id[:8], turn + 1, tool_name, self.task_id[:8], turn + 1, tool_name,
tool_elapsed, pool_active, tool_elapsed, pool_active,
)
except Exception as e:
tool_result = json.dumps(
{"error": f"Tool execution failed: {type(e).__name__}: {str(e)}"}
)
tool_errors.append(ToolError(
turn=turn + 1, tool_name=tool_name,
arguments=tool_args_raw[:200],
error=f"{type(e).__name__}: {str(e)}",
tool_result=tool_result,
))
logger.error(
"Tool '%s' execution failed on turn %d: %s",
tool_name, turn + 1, e,
) )
except Exception as e:
tool_result = json.dumps(
{"error": f"Tool execution failed: {type(e).__name__}: {str(e)}"}
)
tool_errors.append(ToolError(
turn=turn + 1, tool_name=tool_name,
arguments=tool_args_raw[:200],
error=f"{type(e).__name__}: {str(e)}",
tool_result=tool_result,
))
logger.error(
"Tool '%s' execution failed on turn %d: %s",
tool_name, turn + 1, e,
)
# Also check if the tool returned an error in its JSON result # Also check if the tool returned an error in its JSON result
try: try: