wuyang/zk-data-agent
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge pull request #4 from HarnessLab/dev

Browse Source 
add File-history snapshot ids and replay summaries for file edits
This commit is contained in:
Unknown
2026-04-02 22:19:59 +02:00
committed by GitHub
parent 4aae29ae8f 5ad10c660c
commit 3a06266000
5 changed files with 283 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files
PARITY_CHECKLIST.md
+5 -2
View File
@@ -24,12 +24,15 @@ Done:
- [x] File history journaling for write/edit/shell tool actions
- [x] Incremental `bash` tool-result streaming events
- [x] Incremental tool-result streaming for read-only text tools
- [x] Incremental tool-result streaming across the current Python text tool surface
- [x] Mutable tool transcript updates during tool execution
- [x] Transcript mutation history for replaced/tombstoned messages
- [x] Assistant streaming and tool-call transcript mutation history
- [x] Structured transcript block export for messages, tool calls, and tool results
- [x] Resume-time file-history replay reminders
- [x] Resume-time file-history snapshot previews for file edits
- [x] File-history snapshot ids and replay summaries for file edits
- [x] File-history result previews for shell and delegated-tool entries
- [x] Truncated-response continuation flow for `finish_reason=length`
- [x] Basic snipping of older tool/tool-call messages for context control
- [x] Basic automatic compact-boundary insertion with preserved recent tail
@@ -69,11 +72,11 @@ Done:
Missing:
- [ ] Full partial tool-result streaming parity across the complete tool surface
- [ ] Full partial tool-result streaming parity across the complete upstream/npm tool surface
- [ ] Full rich transcript mutation behavior like the npm runtime
- [ ] Full reasoning budgets and task budgets parity
- [ ] Full multi-agent orchestration parity beyond sequential grouped delegation and resumed-child flows
- [ ] Full file history snapshots and replay flows
- [ ] Full file history snapshots and replay flows beyond the current preview/id-based implementation
- [ ] Full executable plugin lifecycle beyond manifest-driven guidance, blocking, aliases, and virtual tools
- [ ] Full session compaction / snipping parity beyond lineage-aware summaries and replay reminders
- [ ] Full `QueryEngine.ts` parity
src/agent_runtime.py
+81 -2
View File
@@ -1159,14 +1159,35 @@ class LocalCodingAgent:
and tool_result.metadata.get('action') != 'delegate_agent'
):
return None
return {
metadata = dict(tool_result.metadata)
entry: dict[str, object] = {
'timestamp': datetime.now(timezone.utc).isoformat(),
'turn_index': turn_index,
'tool_call_id': tool_call.id,
'tool_name': tool_call.name,
'ok': tool_result.ok,
**dict(tool_result.metadata),
'history_entry_id': f'{turn_index}:{tool_call.id}:{tool_call.name}',
'result_preview': self._preview_text(tool_result.content, 220),
**metadata,
}
action = metadata.get('action')
path = metadata.get('path')
if isinstance(path, str) and path:
entry['history_kind'] = 'file_change'
entry['changed_paths'] = [path]
before_sha256 = metadata.get('before_sha256')
if isinstance(before_sha256, str) and before_sha256:
entry['before_snapshot_id'] = f'{path}:{before_sha256[:12]}'
after_sha256 = metadata.get('after_sha256')
if isinstance(after_sha256, str) and after_sha256:
entry['after_snapshot_id'] = f'{path}:{after_sha256[:12]}'
elif isinstance(metadata.get('command'), str):
entry['history_kind'] = 'shell'
elif action == 'delegate_agent':
entry['history_kind'] = 'delegation'
else:
entry['history_kind'] = 'tool'
return entry
def _compact_prefix_count(self, session: AgentSessionState) -> int:
prefix_count = 0
@@ -1711,6 +1732,24 @@ class LocalCodingAgent:
if not file_history:
return
replay_count = len(file_history)
unique_paths = sorted(
{
path
for entry in file_history
for path in (
entry.get('changed_paths')
if isinstance(entry.get('changed_paths'), list)
else ([entry.get('path')] if isinstance(entry.get('path'), str) else [])
)
if isinstance(path, str) and path
}
)
snapshot_count = sum(
1
for entry in file_history
for key in ('before_snapshot_id', 'after_snapshot_id')
if isinstance(entry.get(key), str) and entry.get(key)
)
for message in reversed(session.messages):
if message.metadata.get('kind') != 'file_history_replay':
continue
@@ -1722,6 +1761,8 @@ class LocalCodingAgent:
metadata={
'kind': 'file_history_replay',
'file_history_count': replay_count,
'file_history_unique_paths': len(unique_paths),
'file_history_snapshot_count': snapshot_count,
},
message_id=f'file_history_replay_{replay_count}',
)
@@ -1730,16 +1771,45 @@ class LocalCodingAgent:
self,
file_history: tuple[dict[str, object], ...],
) -> str:
unique_paths = sorted(
{
path
for entry in file_history
for path in (
entry.get('changed_paths')
if isinstance(entry.get('changed_paths'), list)
else ([entry.get('path')] if isinstance(entry.get('path'), str) else [])
)
if isinstance(path, str) and path
}
)
snapshot_count = sum(
1
for entry in file_history
for key in ('before_snapshot_id', 'after_snapshot_id')
if isinstance(entry.get(key), str) and entry.get(key)
)
lines = [
'<system-reminder>',
'Recent file history from this saved session:',
f'- History entries: {len(file_history)}',
f'- Unique changed paths: {len(unique_paths)}',
f'- Snapshot ids: {snapshot_count}',
]
if unique_paths:
preview_paths = ', '.join(unique_paths[:4])
if len(unique_paths) > 4:
preview_paths += f', ... (+{len(unique_paths) - 4} more)'
lines.append(f'- Changed path preview: {preview_paths}')
for entry in file_history[-10:]:
action = str(entry.get('action', entry.get('tool_name', 'tool')))
turn = entry.get('turn_index')
path = entry.get('path')
command = entry.get('command')
details = [f'action={action}']
history_entry_id = entry.get('history_entry_id')
if isinstance(history_entry_id, str) and history_entry_id:
details.append(f'entry_id={history_entry_id}')
if turn is not None:
details.append(f'turn={turn}')
if path:
@@ -1750,12 +1820,21 @@ class LocalCodingAgent:
if isinstance(child_session_ids, list) and child_session_ids:
details.append(f'child_sessions={len(child_session_ids)}')
lines.append(f"- {'; '.join(details)}")
before_snapshot_id = entry.get('before_snapshot_id')
if isinstance(before_snapshot_id, str) and before_snapshot_id:
lines.append(f' before_snapshot: {before_snapshot_id}')
after_snapshot_id = entry.get('after_snapshot_id')
if isinstance(after_snapshot_id, str) and after_snapshot_id:
lines.append(f' after_snapshot: {after_snapshot_id}')
before_preview = entry.get('before_preview')
if isinstance(before_preview, str) and before_preview:
lines.append(f' before: {before_preview}')
after_preview = entry.get('after_preview')
if isinstance(after_preview, str) and after_preview:
lines.append(f' after: {after_preview}')
result_preview = entry.get('result_preview')
if isinstance(result_preview, str) and result_preview:
lines.append(f' result: {result_preview}')
if len(file_history) > 10:
lines.append(f'- ... plus {len(file_history) - 10} older file-history entries')
lines.extend(
src/agent_tools.py
+9 -1
View File
@@ -121,7 +121,7 @@ def execute_tool_streaming(
return
result = tool.execute(arguments, context)
if name in {'list_dir', 'read_file', 'glob_search', 'grep_search'} and result.ok:
if result.ok and result.content and name != 'delegate_agent':
yield from _stream_static_text_result(result)
return
yield ToolStreamUpdate(kind='result', result=result)
@@ -541,6 +541,9 @@ def _run_bash(arguments: dict[str, Any], context: ToolExecutionContext) -> str:
'action': 'bash',
'command': command,
'exit_code': completed.returncode,
'stdout_preview': _snapshot_text(stdout),
'stderr_preview': _snapshot_text(stderr),
'output_preview': _snapshot_text('\n'.join(payload).strip()),
},
)
@@ -634,6 +637,8 @@ def _stream_bash(
'command': command,
'exit_code': exit_code,
'timed_out': True,
'stdout_preview': _snapshot_text(''.join(stdout_chunks)),
'stderr_preview': _snapshot_text(''.join(stderr_chunks)),
},
),
)
@@ -659,6 +664,9 @@ def _stream_bash(
'command': command,
'exit_code': exit_code,
'streamed': True,
'stdout_preview': _snapshot_text(stdout),
'stderr_preview': _snapshot_text(stderr),
'output_preview': _snapshot_text('\n'.join(payload).strip()),
},
),
)
tests/test_agent_runtime.py
+74
View File
@@ -749,7 +749,76 @@ class AgentRuntimeTests(unittest.TestCase):
stored = load_agent_session(result.session_id or '', directory=session_dir)
self.assertEqual(len(result.file_history), 1)
self.assertEqual(result.file_history[0]['path'], 'out.txt')
self.assertEqual(result.file_history[0]['history_kind'], 'file_change')
self.assertIn('history_entry_id', result.file_history[0])
self.assertIn('after_snapshot_id', result.file_history[0])
self.assertEqual(stored.file_history[0]['action'], 'write_file')
self.assertEqual(stored.file_history[0]['after_snapshot_id'], result.file_history[0]['after_snapshot_id'])
def test_agent_streams_write_file_tool_output(self) -> None:
responses = [
{
'choices': [
{
'message': {
'role': 'assistant',
'content': 'Creating the file.',
'tool_calls': [
{
'id': 'call_1',
'type': 'function',
'function': {
'name': 'write_file',
'arguments': '{"path": "out.txt", "content": "hello"}',
},
}
],
},
'finish_reason': 'tool_calls',
}
],
'usage': {'prompt_tokens': 4, 'completion_tokens': 3},
},
{
'choices': [
{
'message': {
'role': 'assistant',
'content': 'Write streamed.',
},
'finish_reason': 'stop',
}
],
'usage': {'prompt_tokens': 5, 'completion_tokens': 2},
},
]
with tempfile.TemporaryDirectory() as tmp_dir:
workspace = Path(tmp_dir)
with patch('src.openai_compat.request.urlopen', side_effect=make_urlopen_side_effect(responses)):
agent = LocalCodingAgent(
model_config=ModelConfig(
model='Qwen/Qwen3-Coder-30B-A3B-Instruct',
base_url='http://127.0.0.1:8000/v1',
),
runtime_config=AgentRuntimeConfig(
cwd=workspace,
permissions=AgentPermissions(allow_file_write=True),
),
)
result = agent.run('Create out.txt')
self.assertEqual(result.final_output, 'Write streamed.')
tool_delta_events = [
event for event in result.events
if event.get('type') == 'tool_delta' and event.get('tool_name') == 'write_file'
]
self.assertGreaterEqual(len(tool_delta_events), 1)
tool_message = next(
message for message in result.transcript
if message.get('role') == 'tool'
)
metadata = tool_message.get('metadata', {})
self.assertEqual(metadata.get('streamed'), True)
self.assertEqual(metadata.get('action'), 'write_file')
def test_agent_streams_bash_tool_output_and_mutates_tool_transcript(self) -> None:
responses = [
@@ -1142,8 +1211,13 @@ class AgentRuntimeTests(unittest.TestCase):
]
self.assertEqual(len(replay_messages), 1)
replay_content = replay_messages[0]['content']
self.assertIn('Unique changed paths: 1', replay_content)
self.assertIn('Snapshot ids: 2', replay_content)
self.assertIn('before_snapshot:', replay_content)
self.assertIn('after_snapshot:', replay_content)
self.assertIn('before: hello world', replay_content)
self.assertIn('after: hello mars', replay_content)
self.assertIn('result:', replay_content)
def test_resume_injects_compaction_replay_reminder(self) -> None:
responses = [
tests/test_query_engine_runtime.py
+114
View File
@@ -1034,3 +1034,117 @@ class QueryEngineRuntimeTests(unittest.TestCase):
self.assertIn('## Runtime Orchestration', summary)
self.assertIn('- group_status:completed=1', summary)
self.assertIn('- child_stop:stop=2', summary)
def test_query_engine_runtime_summary_tracks_resumed_delegate_children(self) -> None:
responses = [
{
'choices': [
{
'message': {
'role': 'assistant',
'content': 'Seed child result.',
},
'finish_reason': 'stop',
}
],
'usage': {'prompt_tokens': 5, 'completion_tokens': 2},
},
{
'choices': [
{
'message': {
'role': 'assistant',
'content': 'Delegating into resumed child.',
'tool_calls': [
{
'id': 'call_1',
'type': 'function',
'function': {
'name': 'delegate_agent',
'arguments': '{}',
},
}
],
},
'finish_reason': 'tool_calls',
}
],
'usage': {'prompt_tokens': 8, 'completion_tokens': 3},
},
{
'choices': [
{
'message': {
'role': 'assistant',
'content': 'Resumed child result.',
},
'finish_reason': 'stop',
}
],
'usage': {'prompt_tokens': 6, 'completion_tokens': 2},
},
{
'choices': [
{
'message': {
'role': 'assistant',
'content': 'Parent completed after resumed child.',
},
'finish_reason': 'stop',
}
],
'usage': {'prompt_tokens': 7, 'completion_tokens': 2},
},
]
with tempfile.TemporaryDirectory() as tmp_dir:
workspace = Path(tmp_dir)
session_dir = workspace / '.port_sessions' / 'agent'
with patch(
'src.openai_compat.request.urlopen',
side_effect=make_urlopen_side_effect(responses),
):
seed_agent = LocalCodingAgent(
model_config=ModelConfig(
model='Qwen/Qwen3-Coder-30B-A3B-Instruct',
base_url='http://127.0.0.1:8000/v1',
),
runtime_config=AgentRuntimeConfig(
cwd=workspace,
session_directory=session_dir,
),
)
seeded = seed_agent.run('Seed the delegated child')
resumed_child_id = seeded.session_id or ''
delegate_arguments = json.dumps(
{
'subtasks': [
{
'label': 'resume_child',
'prompt': 'Continue the delegated child.',
'resume_session_id': resumed_child_id,
'max_turns': 2,
}
],
'max_turns': 2,
}
)
responses[1]['choices'][0]['message']['tool_calls'][0]['function']['arguments'] = delegate_arguments
agent = LocalCodingAgent(
model_config=ModelConfig(
model='Qwen/Qwen3-Coder-30B-A3B-Instruct',
base_url='http://127.0.0.1:8000/v1',
),
runtime_config=AgentRuntimeConfig(
cwd=workspace,
session_directory=session_dir,
),
)
engine = QueryEnginePort.from_runtime_agent(agent)
turn = engine.submit_message('Delegate into resumed child')
summary = engine.render_summary()
self.assertEqual(turn.output, 'Parent completed after resumed child.')
self.assertIn('## Runtime Orchestration', summary)
self.assertIn('- resumed_children=1', summary)