feat(plugins): let pre_tool_call hooks block tool execution

Plugins can now return {"action": "block", "message": "reason"} from
their pre_tool_call hook to prevent a tool from executing. The error
message is returned to the model as a tool result so it can adjust.

Covers both execution paths: handle_function_call (model_tools.py) and
agent-level tools (run_agent.py _invoke_tool + sequential/concurrent).
Blocked tools skip all side effects (counter resets, checkpoints,
callbacks, read-loop tracker).

Adds skip_pre_tool_call_hook flag to avoid double-firing the hook when
run_agent.py already checked and then calls handle_function_call.

Salvaged from PR #5385 (gianfrancopiana) and PR #4610 (oredsecurity).
This commit is contained in:
Gianfranco Piana 2026-04-13 21:15:25 -07:00 committed by Teknium
parent ea74f61d98
commit eabc0a2f66
6 changed files with 335 additions and 40 deletions

View file

@ -584,6 +584,45 @@ def invoke_hook(hook_name: str, **kwargs: Any) -> List[Any]:
def get_pre_tool_call_block_message(
tool_name: str,
args: Optional[Dict[str, Any]],
task_id: str = "",
session_id: str = "",
tool_call_id: str = "",
) -> Optional[str]:
"""Check ``pre_tool_call`` hooks for a blocking directive.
Plugins that need to enforce policy (rate limiting, security
restrictions, approval workflows) can return::
{"action": "block", "message": "Reason the tool was blocked"}
from their ``pre_tool_call`` callback. The first valid block
directive wins. Invalid or irrelevant hook return values are
silently ignored so existing observer-only hooks are unaffected.
"""
hook_results = invoke_hook(
"pre_tool_call",
tool_name=tool_name,
args=args if isinstance(args, dict) else {},
task_id=task_id,
session_id=session_id,
tool_call_id=tool_call_id,
)
for result in hook_results:
if not isinstance(result, dict):
continue
if result.get("action") != "block":
continue
message = result.get("message")
if isinstance(message, str) and message:
return message
return None
def get_plugin_context_engine():
"""Return the plugin-registered context engine, or None."""
return get_plugin_manager()._context_engine