feat: harden video analysis and runtime resilience

This commit is contained in:
somew1nd committed 2026-08-25 20:58:25 -07:00
1 parent 5ada543a09
commit bb16991f2c
55 files changed
+5023 -437

No files matched your search

@@ -0,0 +1,93 @@
import { consolidateLogsToBlocks, getSortedStepEvents } from './stream-aggregator.util';
describe('stream aggregator timeline ordering', () => {
it('interleaves text, tools, and actions by timestamp', () => {
const events = getSortedStepEvents({
timestamp: 10,
operator_native_thinking: 'Thought content',
operator_native_thinking_timestamp: 11,
operator_raw_thinking: 'Work content',
operator_raw_thinking_timestamp: 13,
action_taken: { action: 'tap', timestamp: 15 },
generic_tools: [
{ trace_id: 'tool-late', name: 'save_note', timestamp: 14 },
{ trace_id: 'tool-early', name: 'video_analyzer', timestamp: 12 }
]
});
expect(events.map((event) => event.type)).toEqual([
'thinking',
'tool',
'text',
'tool',
'action'
]);
expect(events.map((event) => event.timestamp)).toEqual([
11000,
12000,
13000,
14000,
15000
]);
});
it('uses a deterministic text-first fallback when only a step timestamp exists', () => {
const events = getSortedStepEvents({
timestamp: 20,
operator_native_thinking: 'Thought content',
operator_raw_thinking: 'Work content',
action_taken: { action: 'tap' },
generic_tools: [{ trace_id: 'tool-1', name: 'save_note' }]
});
expect(events.map((event) => event.type)).toEqual([
'thinking',
'text',
'action',
'tool'
]);
});
it('keeps the first live timestamp when later chunks update the same text stream', () => {
const blocks = consolidateLogsToBlocks([
{
type: 'llm_stream',
timestamp: '2026-08-25T20:00:01.000Z',
data: {
execution_id: 'exec-1',
step_id: 'step-1',
stream_type: 'text',
text: 'First',
isCompleted: false
}
},
{
type: 'trace_recorded',
timestamp: '2026-08-25T20:00:02.000Z',
data: {
trace_id: 'tool-1',
step_id: 'step-1',
type: 'tool',
name: 'save_note',
timestamp: '2026-08-25T20:00:02.000Z'
}
},
{
type: 'llm_stream',
timestamp: '2026-08-25T20:00:03.000Z',
data: {
execution_id: 'exec-1',
step_id: 'step-1',
stream_type: 'text',
text: 'First and second',
isCompleted: false
}
}
]);
expect(blocks[0].data.operator_raw_thinking_timestamp)
.toBe('2026-08-25T20:00:01.000Z');
expect(getSortedStepEvents(blocks[0].data).map((event) => event.type))
.toEqual(['text', 'tool']);
});
});
@@ -64,8 +64,12 @@ export function consolidateLogsToBlocks(rawLogs: any[]): StepBlock[] {
const updatedData = { ...existing.data };
if (streamType === 'thinking') {
updatedData.operator_native_thinking = text;
updatedData.operator_native_thinking_timestamp =
updatedData.operator_native_thinking_timestamp || log.timestamp;
} else {
updatedData.operator_raw_thinking = text;
updatedData.operator_raw_thinking_timestamp =
updatedData.operator_raw_thinking_timestamp || log.timestamp;
}
if (stepId && !updatedData.step_id) {
updatedData.step_id = stepId;
@@ -89,8 +93,10 @@ export function consolidateLogsToBlocks(rawLogs: any[]): StepBlock[] {
};
if (streamType === 'thinking') {
blockData.operator_native_thinking = text;
blockData.operator_native_thinking_timestamp = log.timestamp;
} else {
blockData.operator_raw_thinking = text;
blockData.operator_raw_thinking_timestamp = log.timestamp;
}
blocks.push({
@@ -134,6 +140,12 @@ export function consolidateLogsToBlocks(rawLogs: any[]): StepBlock[] {
...log.data,
operator_native_thinking: log.data.operator_native_thinking || existingBlock.data.operator_native_thinking || undefined,
operator_raw_thinking: log.data.operator_raw_thinking || existingBlock.data.operator_raw_thinking || undefined,
operator_native_thinking_timestamp: existingBlock.data.operator_native_thinking_timestamp
|| log.data.operator_native_thinking_timestamp
|| (log.data.operator_native_thinking ? log.timestamp : undefined),
operator_raw_thinking_timestamp: existingBlock.data.operator_raw_thinking_timestamp
|| log.data.operator_raw_thinking_timestamp
|| (log.data.operator_raw_thinking ? log.timestamp : undefined),
action_taken: log.data.action_taken || existingBlock.data.action_taken || undefined,
last_execution_result: log.data.last_execution_result || existingBlock.data.last_execution_result || undefined,
generic_tools: mergedTools,
@@ -155,6 +167,10 @@ export function consolidateLogsToBlocks(rawLogs: any[]): StepBlock[] {
data: {
...log.data,
isCompleted: true,
operator_native_thinking_timestamp: log.data.operator_native_thinking_timestamp
|| (log.data.operator_native_thinking ? log.timestamp : undefined),
operator_raw_thinking_timestamp: log.data.operator_raw_thinking_timestamp
|| (log.data.operator_raw_thinking ? log.timestamp : undefined),
generic_tools: log.data.generic_tools || []
}
});
@@ -410,23 +426,58 @@ export function groupBlocksToPhases(blocks: StepBlock[], sessionStartTime: numbe
*/
export function getSortedStepEvents(
stepData: any,
cache?: WeakMap<any, { length: number; actionTs: any; events: StepEvent[] }>
cache?: WeakMap<any, { signature: string; events: StepEvent[] }>
): StepEvent[] {
if (!stepData || typeof stepData !== 'object') return [];
const toolsLen = stepData.generic_tools ? stepData.generic_tools.length : 0;
const actionTs = stepData.action_taken ? (stepData.action_taken.timestamp || stepData.action_taken.start_time || stepData.action_taken.created_at) : null;
const nativeText = typeof stepData.operator_native_thinking === 'string'
? stepData.operator_native_thinking.trim()
: '';
const rawText = typeof stepData.operator_raw_thinking === 'string'
? stepData.operator_raw_thinking.trim()
: '';
const signature = [
toolsLen,
actionTs ?? '',
stepData.operator_native_thinking_timestamp ?? '',
stepData.operator_raw_thinking_timestamp ?? '',
nativeText.length,
rawText.length
].join('|');
if (cache) {
const cached = cache.get(stepData);
if (cached && cached.length === toolsLen && cached.actionTs === actionTs) {
if (cached?.signature === signature) {
return cached.events;
}
}
const events: Array<{ type: 'action' | 'tool'; data: any; timestamp: number }> = [];
const events: Array<{ type: StepEvent['type']; data: any; timestamp: number; sequence: number }> = [];
const fallbackTime = getItemTimestamp(stepData.timestamp, 0);
let sequence = 0;
// 1. Add action_taken if present and valid
// Text and tools share one timeline. Live streams carry their first-seen
// timestamps; persisted steps use the step timestamp as a stable fallback.
if (nativeText) {
events.push({
type: 'thinking',
data: { text: stepData.operator_native_thinking },
timestamp: getItemTimestamp(stepData.operator_native_thinking_timestamp, fallbackTime),
sequence: sequence++
});
}
if (rawText) {
events.push({
type: 'text',
data: { text: stepData.operator_raw_thinking },
timestamp: getItemTimestamp(stepData.operator_raw_thinking_timestamp, fallbackTime),
sequence: sequence++
});
}
// Add action_taken if present and valid.
if (stepData.action_taken && (isAndroidAction(stepData.action_taken) || isReportStatusAction(stepData.action_taken))) {
const actTime = getItemTimestamp(
stepData.action_taken.timestamp ?? stepData.action_taken.start_time ?? stepData.action_taken.created_at,
@@ -435,11 +486,12 @@ export function getSortedStepEvents(
events.push({
type: 'action',
data: stepData.action_taken,
timestamp: actTime
timestamp: actTime,
sequence: sequence++
});
}
// 2. Add generic tools if present
// Add generic tools if present.
const uniqueTools = getUniqueGenericTools(stepData.generic_tools);
uniqueTools.forEach((tool: any) => {
const toolTime = getItemTimestamp(
@@ -449,16 +501,22 @@ export function getSortedStepEvents(
events.push({
type: 'tool',
data: tool,
timestamp: toolTime
timestamp: toolTime,
sequence: sequence++
});
});
// 3. Sort chronologically by timestamp
events.sort((a, b) => a.timestamp - b.timestamp);
// Sort chronologically. The explicit sequence makes equal/missing timestamp
// behavior deterministic across browsers and preserves the legacy fallback.
events.sort((a, b) => (a.timestamp - b.timestamp) || (a.sequence - b.sequence));
const result: StepEvent[] = events.map(e => ({ type: e.type, data: e.data }));
const result: StepEvent[] = events.map(e => ({
type: e.type,
data: e.data,
timestamp: e.timestamp
}));
if (cache) {
cache.set(stepData, { length: toolsLen, actionTs, events: result });
cache.set(stepData, { signature, events: result });
}
return result;
}
@@ -1,4 +1,8 @@
import { cleanErrorMessage, getUniqueGenericTools } from './tool-formatter.util';
import {
cleanErrorMessage,
getUniqueGenericTools,
getVideoAnalysisView
} from './tool-formatter.util';
describe('cleanErrorMessage', () => {
it('removes repeated LLM wrapper labels while preserving the provider reason', () => {
@@ -86,3 +90,77 @@ describe('getUniqueGenericTools retry aggregation', () => {
expect(result.map(tool => tool.name)).toEqual(['llm_pause']);
});
});
describe('video analysis timeline formatting', () => {
it('distinguishes a cache hit without presenting it as running', () => {
const view = getVideoAnalysisView({
name: 'spawn_sub_agent',
status: 'success',
payload: {
args: { start_time: 0, end_time: 42 },
result: 'CACHED VIDEO ANALYSIS: existing evidence'
}
});
expect(view?.outcome).toBe('complete');
expect(view?.reuse).toBe('full');
expect(view?.title).toBe('Reused video analysis');
expect(view?.requestedRange).toEqual({ start: 0, end: 42 });
});
it('collapses child video chunks into one stable note-style timeline item', () => {
const result = getUniqueGenericTools([
{
trace_id: 'chunk-1',
parent_trace_id: 'video-agent-1',
type: 'tool',
name: 'spawn_sub_agent',
status: 'success',
timestamp: 100,
payload: {
args: { start_time: 0, end_time: 30, specific_query: 'find the result' },
result: '[from 0.0s to 30.0s] Summary: first'
}
},
{
trace_id: 'chunk-2',
parent_trace_id: 'video-agent-1',
type: 'tool',
name: 'spawn_sub_agent',
status: 'success',
timestamp: 101,
payload: {
args: { start_time: 30, end_time: 60, specific_query: 'find the result' },
result: 'PARTIAL VIDEO ANALYSIS (successful chunks were persisted). Failed intervals: 45.0s-60.0s'
}
}
]);
expect(result.length).toBe(1);
expect(result[0].name).toBe('video_analysis');
expect(result[0].trace_id).toBe('video-analysis-video-agent-1');
expect(result[0].payload.result.outcome).toBe('partial');
expect(result[0].payload.result.requested_range).toEqual({ start: 0, end: 60 });
expect(getVideoAnalysisView(result[0])?.title).toBe('Video analysis partially completed');
});
it('keeps video analyses from independent parent executions separate', () => {
const result = getUniqueGenericTools([
{
trace_id: 'chunk-1', parent_trace_id: 'video-agent-1', type: 'tool',
name: 'spawn_sub_agent', status: 'success',
payload: { args: { start_time: 0, end_time: 10 }, result: 'first' }
},
{
trace_id: 'chunk-2', parent_trace_id: 'video-agent-2', type: 'tool',
name: 'spawn_sub_agent', status: 'success',
payload: { args: { start_time: 10, end_time: 20 }, result: 'second' }
}
]);
expect(result.map(tool => tool.trace_id)).toEqual([
'video-analysis-video-agent-1',
'video-analysis-video-agent-2'
]);
});
});
@@ -64,7 +64,133 @@ export function isVideoTool(tool: any): boolean {
if (!tool || !tool.name) return false;
const cleanName = tool.name.replace(/^(_)?(self\.)?exec_/, '');
const nameLower = cleanName.toLowerCase();
return ['video_analyzer', 'video_analyzer_pure'].includes(nameLower);
return [
'video_analysis',
'video_analyzer',
'video_analyzer_pure',
'spawn_sub_agent',
'analyze_audio_only'
].includes(nameLower);
}
export type VideoAnalysisOutcome = 'running' | 'recovering' | 'waiting' | 'complete' | 'partial' | 'failed';
export interface VideoAnalysisRange {
start: number;
end: number;
category?: string;
retryable?: boolean;
}
export interface VideoAnalysisView {
outcome: VideoAnalysisOutcome;
title: string;
query: string;
summary: string;
reuse: 'none' | 'partial' | 'full';
requestedRange: VideoAnalysisRange | null;
completedRanges: VideoAnalysisRange[];
failedRanges: VideoAnalysisRange[];
evidenceCount: number;
completedCount: number;
totalCount: number;
fallbackUsed: boolean;
}
function numericRange(value: any): VideoAnalysisRange | null {
if (!value || typeof value !== 'object') return null;
const start = Number(value.start);
const end = Number(value.end);
if (!Number.isFinite(start) || !Number.isFinite(end) || end <= start) return null;
return {
start,
end,
category: value.category ? String(value.category) : undefined,
retryable: value.retryable === true
};
}
function parseVideoResultText(text: string): Partial<VideoAnalysisView> {
const value = text.trim();
if (!value) return {};
if (value.startsWith('CACHED VIDEO ANALYSIS:')) {
return { outcome: 'complete', reuse: 'full', summary: value.replace(/^CACHED VIDEO ANALYSIS:\s*/, '') };
}
if (value.startsWith('PARTIAL VIDEO ANALYSIS')) {
return { outcome: 'partial', reuse: 'partial', summary: value };
}
if (value.startsWith('All sub-agent chunks failed') || value.startsWith('Error:')) {
return { outcome: 'failed', summary: value };
}
if (value.includes('Analysis is already in progress in another video agent')) {
return { outcome: 'waiting', summary: 'Another video agent is already analyzing this evidence.' };
}
return { outcome: 'complete', summary: value };
}
/** Build the user-facing, backward-compatible video analysis state. */
export function getVideoAnalysisView(tool: any): VideoAnalysisView | null {
if (!isVideoTool(tool)) return null;
const args = getToolArgs(tool);
const payload = tool?.payload && typeof tool.payload === 'object' ? tool.payload : {};
const rawResult = payload.result ?? tool.result ?? null;
const structured = rawResult && typeof rawResult === 'object' ? rawResult : {};
const parsed = typeof rawResult === 'string' ? parseVideoResultText(rawResult) : {};
let outcome = String(structured.outcome || payload.outcome || parsed.outcome || '').toLowerCase() as VideoAnalysisOutcome;
if (!['running', 'recovering', 'waiting', 'complete', 'partial', 'failed'].includes(outcome)) {
outcome = tool?.status === 'failed' || tool?.status === 'error'
? 'failed'
: tool?.status === 'running' ? 'running' : 'complete';
}
const recovering = structured.recovering === true || structured.fallback_used === true;
if (outcome === 'running' && recovering) outcome = 'recovering';
const start = Number(args.start_time ?? structured.requested_range?.start);
const end = Number(args.end_time ?? structured.requested_range?.end);
const requestedRange = Number.isFinite(start) && Number.isFinite(end) && end > start
? { start, end }
: numericRange(structured.requested_range);
const completedRanges = Array.isArray(structured.completed_ranges)
? structured.completed_ranges.map(numericRange).filter((range: VideoAnalysisRange | null): range is VideoAnalysisRange => Boolean(range))
: [];
const failedRanges = Array.isArray(structured.failed_ranges)
? structured.failed_ranges.map(numericRange).filter((range: VideoAnalysisRange | null): range is VideoAnalysisRange => Boolean(range))
: [];
const completedCount = Number(structured.completed_count ?? completedRanges.length ?? 0);
const totalCount = Number(structured.total_count ?? (completedRanges.length + failedRanges.length));
const titleByOutcome: Record<VideoAnalysisOutcome, string> = {
running: 'Analyzing screen recording',
recovering: 'Analyzing unfinished recording segment',
waiting: 'Waiting for existing video analysis',
complete: structured.reuse === 'full' || parsed.reuse === 'full'
? 'Reused video analysis'
: 'Analyzed screen recording',
partial: 'Video analysis partially completed',
failed: 'Video analysis returned no result'
};
return {
outcome,
title: titleByOutcome[outcome],
query: String(args.specific_query || args.query || args.prompt || structured.query || ''),
summary: String(structured.summary || parsed.summary || ''),
reuse: (structured.reuse || parsed.reuse || 'none') as 'none' | 'partial' | 'full',
requestedRange,
completedRanges,
failedRanges,
evidenceCount: Number(structured.evidence_count || 0),
completedCount: Number.isFinite(completedCount) ? completedCount : 0,
totalCount: Number.isFinite(totalCount) ? totalCount : 0,
fallbackUsed: structured.fallback_used === true
};
}
export function formatVideoTime(seconds: number): string {
const safe = Math.max(0, Number(seconds) || 0);
const minutes = Math.floor(safe / 60);
const secs = Math.floor(safe % 60);
return `${String(minutes).padStart(2, '0')}:${String(secs).padStart(2, '0')}`;
}
/**
@@ -301,7 +427,107 @@ export function getUniqueGenericTools(tools: any[] | undefined): any[] {
}
}
}
return result;
return collapseVideoAnalysisTools(result);
}
/**
* Present one note-style timeline row for one video-analyzer execution. Child
* chunk, audio, and wrapper traces stay available in the raw trace but do not
* look like repeated user-visible analyses.
*/
function collapseVideoAnalysisTools(tools: any[]): any[] {
const groups = new Map<string, { firstIndex: number; tools: any[] }>();
const passthrough: Array<{ index: number; tool: any }> = [];
tools.forEach((tool, index) => {
if (!isVideoTool(tool)) {
passthrough.push({ index, tool });
return;
}
const cleanName = String(tool.name || '').replace(/^(_)?(self\.)?exec_/, '').toLowerCase();
const groupId = cleanName === 'video_analyzer' || cleanName === 'video_analyzer_pure'
? String(tool.trace_id || tool.parent_trace_id || `video-${index}`)
: String(tool.parent_trace_id || tool.trace_id || `video-${index}`);
const existing = groups.get(groupId);
if (existing) {
existing.tools.push(tool);
} else {
groups.set(groupId, { firstIndex: index, tools: [tool] });
}
});
const collapsed = [...passthrough];
for (const [groupId, group] of groups) {
const views = group.tools
.map(getVideoAnalysisView)
.filter((view): view is VideoAnalysisView => Boolean(view));
if (!views.length) continue;
const hasActive = views.some(view => view.outcome === 'running' || view.outcome === 'recovering');
const completed = views.filter(view => view.outcome === 'complete');
const hasPartial = views.some(view => view.outcome === 'partial');
const failed = views.filter(view => view.outcome === 'failed');
const waiting = views.filter(view => view.outcome === 'waiting');
let outcome: VideoAnalysisOutcome;
if (hasActive) {
outcome = views.some(view => view.outcome === 'recovering') ? 'recovering' : 'running';
} else if (hasPartial || (completed.length > 0 && failed.length > 0)) {
outcome = 'partial';
} else if (completed.length > 0) {
outcome = 'complete';
} else if (waiting.length > 0) {
outcome = 'waiting';
} else {
outcome = 'failed';
}
const ranges = views
.map(view => view.requestedRange)
.filter((range): range is VideoAnalysisRange => Boolean(range));
const requestedRange = ranges.length
? {
start: Math.min(...ranges.map(range => range.start)),
end: Math.max(...ranges.map(range => range.end))
}
: null;
const reuse = completed.length > 0 && completed.every(view => view.reuse === 'full')
? 'full'
: views.some(view => view.reuse !== 'none') ? 'partial' : 'none';
const base = group.tools[0];
const structuredResult = {
outcome,
reuse,
requested_range: requestedRange,
completed_count: completed.length,
total_count: views.length,
evidence_count: views.reduce((sum, view) => sum + view.evidenceCount, 0),
query: views.find(view => view.query)?.query || '',
recovering: outcome === 'recovering',
fallback_used: views.some(view => view.fallbackUsed)
};
collapsed.push({
index: group.firstIndex,
tool: {
...base,
trace_id: `video-analysis-${groupId}`,
name: 'video_analysis',
status: outcome === 'running' || outcome === 'recovering' ? 'running' : 'success',
payload: {
...(base.payload || {}),
args: {
...(base.payload?.args || base.args || {}),
start_time: requestedRange?.start,
end_time: requestedRange?.end
},
result: structuredResult
}
}
});
}
collapsed.sort((a, b) => a.index - b.index);
return collapsed.map(item => item.tool);
}
/**