mirrors/hermes-agent
1
0
Fork 0
mirror of https://github.com/NousResearch/hermes-agent.git synced 2026-05-14 04:02:26 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 6

fix(run_agent): break permanent empty-response loop from orphan tool-tail (#21385)

Browse source 
When empty-response terminal scaffolding fires on a tool-result turn,
_drop_trailing_empty_response_scaffolding left the live history ending at
a bare 'tool' message. The next user input then landed as [...tool, user],
a protocol-invalid sequence that OpenRouter/Opus and other providers
silently fail on (returns empty content). That retriggered the empty-retry
recovery every turn, and recovery flags never hit SQLite (no column for
them), so history kept looking broken on every reload.

Two fixes:

1. Scaffolding strip rewinds the orphan assistant(tool_calls)+tool pair
   after popping sentinels. Only fires when scaffolding flags were
   actually present, so mid-iteration tool loops are untouched.

2. _repair_message_sequence runs right before every API call as a
   defensive belt: drops stray tool messages with unknown tool_call_ids,
   merges consecutive user messages so no user input is lost. Does NOT
   rewind assistant(tool_calls)+tool+user — that pattern is valid when
   the user redirected before the model got its continuation turn.

Repro: session 20260507_044111_fa7e65. Opus-4.7/OpenRouter returned
content-less response after a 42KB execute_code output, nudge+retry
chain exhausted (no fallback configured), terminal sentinel appended,
scaffolding stripped leaving bare tool tail, user typed 'wtf happened..'
and landed as tool→user violation. Every subsequent turn collapsed in
<50ms with the same 3-retry empty chain because the API request itself
was malformed.

Verified live via HTTP mock: pre-fix reproduced 5 api_calls/0.15s exit
'empty_response_exhausted'; post-fix 1 api_call/0.10s exit
'text_response(finish_reason=stop)'. Three-turn session flows cleanly
through the scenario. Full run_agent suite: 1242 passed (0 regressions,
2 pre-existing concurrent_interrupt failures unrelated).
This commit is contained in:
Teknium 2026-05-07 08:35:10 -07:00 • committed by GitHub
parent 1d2029b2b7
commit 812ce0b987
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
3 changed files with 373 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

16
tests/run_agent/test_empty_response_recovery_persistence.py
View file

@ -21,9 +21,21 @@ def _agent_with_stubbed_persistence():
def test_persist_session_strips_trailing_empty_recovery_scaffolding():
"""After stripping scaffolding, also rewind past orphan trailing tool-result
messages that the failed iteration left behind. Otherwise the next user
message lands after a bare ``tool`` and produces a protocol-invalid
sequence that most providers silently fail on, retriggering the empty-
retry loop indefinitely.
"""
agent = _agent_with_stubbed_persistence()
messages = [
{"role": "user", "content": "run the task"},
{
"role": "assistant",
"content": "",
"tool_calls": [{"id": "call_1", "type": "function",
"function": {"name": "x", "arguments": "{}"}}],
},
{"role": "tool", "content": "{}", "tool_call_id": "call_1"},
{
"role": "assistant",
@ -42,9 +54,11 @@ def test_persist_session_strips_trailing_empty_recovery_scaffolding():
AIAgent._persist_session(agent, messages, conversation_history=[])
# After strip + rewind, only the original user message remains. The
# assistant(tool_calls) + tool pair is dropped because its iteration
# never produced a real response.
assert messages == [
{"role": "user", "content": "run the task"},
{"role": "tool", "content": "{}", "tool_call_id": "call_1"},
]
assert agent.saved_session_logs[-1] == messages
assert all(not msg.get("_empty_recovery_synthetic") for msg in messages)