apunkt/SurfSense
1
0
Fork 0
mirror of https://github.com/MODSetter/SurfSense.git synced 2026-06-06 20:15:17 +02:00
Code Issues Projects Releases Packages Wiki Activity

docs(agents): refresh comments that referenced the deleted single-agent stack (bucket B6)

Browse source 
After deleting app/agents/new_chat/, several shared-kernel comments still cited
new_chat paths/cycles. Update the two lazy-import comments in middleware to state
the real reason (tools.registry <-> shared.middleware cycle), and repoint dangling
``new_chat/tools/hitl.py`` / ``chat_deepagent`` doc references to their shared
locations. Comment-only; suite unaffected.
This commit is contained in:
CREDO23 2026-06-04 13:47:10 +02:00
parent 305a8fe7e6
commit 8faa03889d
6 changed files with 16 additions and 16 deletions
Download patch file Download diff file Expand all files Collapse all files

6
surfsense_backend/app/agents/shared/middleware/action_log.py
View file

@ -39,9 +39,9 @@ if TYPE_CHECKING: # pragma: no cover - type-only
from langchain.agents.middleware.types import ToolCallRequest
from langgraph.types import Command
# Type-only import: keeping it lazy avoids a module-load cycle through the
# frozen single-agent package (new_chat.__init__ -> chat_deepagent ->
# middleware shim). Resolves to app.agents.shared.tools once tools migrate.
# Type-only import: ToolDefinition is only referenced in annotations, and a
# runtime import would close a module-load cycle (tools.registry imports
# shared.middleware.dedup_tool_calls).
from app.agents.shared.tools.registry import ToolDefinition

2
surfsense_backend/app/agents/shared/middleware/doom_loop.py
View file

@ -16,7 +16,7 @@ This ships **OFF by default** until the frontend explicitly handles
``context.permission == "doom_loop"`` interrupts.
Wire format: uses SurfSense's existing ``interrupt()`` payload shape
(see ``app/agents/new_chat/tools/hitl.py``):
(see ``app/agents/shared/tools/hitl.py``):
{
"type": "permission_ask",

4
surfsense_backend/app/agents/shared/middleware/permission.py
View file

@ -8,7 +8,7 @@ per-thread overrides, and no auto-deny synthesis.
This middleware ports OpenCode's ``packages/opencode/src/permission/index.ts``
ruleset model on top of SurfSense's existing ``interrupt({type, action,
context})`` payload shape (see ``app/agents/new_chat/tools/hitl.py``) so
context})`` payload shape (see ``app/agents/shared/tools/hitl.py``) so
the frontend keeps working unchanged.
Operation:
@ -85,7 +85,7 @@ def _default_pattern_resolver(name: str) -> PatternResolver:
# Translation from the LangChain HITL envelope (what ``stream_resume_chat``
# sends) to SurfSense's legacy ``decision_type`` shape. ``edit`` keeps the
# original tool args — tools needing argument edits should use
# ``request_approval`` from ``app/agents/new_chat/tools/hitl.py``.
# ``request_approval`` from ``app/agents/shared/tools/hitl.py``.
_LC_TYPE_TO_PERMISSION_DECISION: dict[str, str] = {
"approve": "once",
"reject": "reject",

6
surfsense_backend/app/agents/shared/middleware/tool_call_repair.py
View file

@ -118,9 +118,9 @@ class ToolCallNameRepairMiddleware(
return call
# Stage 2 — invalid fallback
# Local import avoids a module-load cycle through the frozen single-agent
# package (new_chat.__init__ -> chat_deepagent -> middleware shim).
# Resolves to app.agents.shared.tools once tools migrate.
# Local import avoids a module-load cycle: tools.registry imports
# shared.middleware (dedup_tool_calls), so importing tools at module
# scope here would close the loop.
from app.agents.shared.tools.invalid_tool import INVALID_TOOL_NAME
if INVALID_TOOL_NAME in registered: