feat(bf-52d6): add conversation transcript parser module
Add standalone conversationParser module to extract full conversation transcripts from NEEDLE logs. This completes Phase 1 Core Infrastructure item for conversation parsing. Implementation: - isConversationSpanEvent(): Identify conversation-related span events - buildConversationSessions(): Build sessions from log events - getWorkerConversationSessions(): Filter sessions by worker - getBeadConversationSession(): Get session for a specific bead - extractConversationEvents(): Extract all conversation events Types supported: - PromptEvent: User input/prompt - ResponseEvent: Assistant response text - ThinkingEvent: Internal reasoning/thinking blocks - ToolCallEvent: Tool invocations with arguments - ToolResultEvent: Tool call results Also updates docs/plan.md to mark Phase 1 Core Infrastructure items complete. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
This commit is contained in:
jedarden
parent
1484adb7a1
commit
3da90f1be0
3 changed files with 1045 additions and 5 deletions
10
docs/plan.md
10
docs/plan.md
|
|
@ -1341,11 +1341,11 @@ fabric logs --worker w-abc123 # Filter by worker
|
|||
## Implementation Phases
|
||||
|
||||
### Phase 1: Core Infrastructure
|
||||
- [ ] Log tailer that watches `~/.needle/logs/`
|
||||
- [ ] JSON line parser
|
||||
- [ ] Event emitter for parsed events
|
||||
- [ ] In-memory event index (by worker, bead, file, timestamp)
|
||||
- [ ] Conversation transcript parser (extract full conversation from logs)
|
||||
- [x] Log tailer that watches `~/.needle/logs/`
|
||||
- [x] JSON line parser
|
||||
- [x] Event emitter for parsed events
|
||||
- [x] In-memory event index (by worker, bead, file, timestamp)
|
||||
- [x] Conversation transcript parser (extract full conversation from logs)
|
||||
|
||||
### Phase 2: TUI Display
|
||||
- [ ] Worker list panel
|
||||
|
|
|
|||
474
src/conversationParser.test.ts
Normal file
474
src/conversationParser.test.ts
Normal file
|
|
@ -0,0 +1,474 @@
|
|||
/**
|
||||
* Tests for conversationParser module
|
||||
*/
|
||||
|
||||
import { describe, it, expect } from 'vitest';
|
||||
import {
|
||||
isConversationSpanEvent,
|
||||
buildConversationSessions,
|
||||
getWorkerConversationSessions,
|
||||
getBeadConversationSession,
|
||||
extractConversationEvents,
|
||||
} from './conversationParser.js';
|
||||
import { LogEvent } from './types.js';
|
||||
|
||||
describe('conversationParser', () => {
|
||||
describe('isConversationSpanEvent', () => {
|
||||
it('should identify llm.request span events', () => {
|
||||
const event: LogEvent = {
|
||||
ts: Date.now(),
|
||||
worker: 'tcb-alpha',
|
||||
sequence: 1,
|
||||
level: 'info',
|
||||
msg: 'llm.request.started',
|
||||
span_name: 'llm.request',
|
||||
};
|
||||
expect(isConversationSpanEvent(event)).toBe(true);
|
||||
});
|
||||
|
||||
it('should identify tool.call span events', () => {
|
||||
const event: LogEvent = {
|
||||
ts: Date.now(),
|
||||
worker: 'tcb-alpha',
|
||||
sequence: 1,
|
||||
level: 'info',
|
||||
msg: 'tool.call.started',
|
||||
span_name: 'tool.call',
|
||||
tool: 'Read',
|
||||
};
|
||||
expect(isConversationSpanEvent(event)).toBe(true);
|
||||
});
|
||||
|
||||
it('should identify events with prompt field', () => {
|
||||
const event: LogEvent = {
|
||||
ts: Date.now(),
|
||||
worker: 'tcb-alpha',
|
||||
sequence: 1,
|
||||
level: 'info',
|
||||
msg: 'bead.prompt_built',
|
||||
prompt: 'Help me write code',
|
||||
};
|
||||
expect(isConversationSpanEvent(event)).toBe(true);
|
||||
});
|
||||
|
||||
it('should identify events with response field', () => {
|
||||
const event: LogEvent = {
|
||||
ts: Date.now(),
|
||||
worker: 'tcb-alpha',
|
||||
sequence: 1,
|
||||
level: 'info',
|
||||
msg: 'agent.response',
|
||||
response: 'I can help with that',
|
||||
};
|
||||
expect(isConversationSpanEvent(event)).toBe(true);
|
||||
});
|
||||
|
||||
it('should return false for non-conversation events', () => {
|
||||
const event: LogEvent = {
|
||||
ts: Date.now(),
|
||||
worker: 'tcb-alpha',
|
||||
sequence: 1,
|
||||
level: 'info',
|
||||
msg: 'worker.state_transition',
|
||||
};
|
||||
expect(isConversationSpanEvent(event)).toBe(false);
|
||||
});
|
||||
|
||||
it('should identify bead.agent_started events', () => {
|
||||
const event: LogEvent = {
|
||||
ts: Date.now(),
|
||||
worker: 'tcb-alpha',
|
||||
sequence: 1,
|
||||
level: 'info',
|
||||
msg: 'bead.agent_started',
|
||||
bead: 'bd-test',
|
||||
};
|
||||
expect(isConversationSpanEvent(event)).toBe(true);
|
||||
});
|
||||
|
||||
it('should identify bead.agent_completed events', () => {
|
||||
const event: LogEvent = {
|
||||
ts: Date.now(),
|
||||
worker: 'tcb-alpha',
|
||||
sequence: 1,
|
||||
level: 'info',
|
||||
msg: 'bead.agent_completed',
|
||||
bead: 'bd-test',
|
||||
};
|
||||
expect(isConversationSpanEvent(event)).toBe(true);
|
||||
});
|
||||
});
|
||||
|
||||
describe('buildConversationSessions', () => {
|
||||
it('should build a session from llm.request events', () => {
|
||||
const events: LogEvent[] = [
|
||||
{
|
||||
ts: Date.now() - 1000,
|
||||
worker: 'tcb-alpha',
|
||||
sequence: 1,
|
||||
level: 'info',
|
||||
msg: 'llm.request.started',
|
||||
span_name: 'llm.request',
|
||||
bead: 'bd-test',
|
||||
session: 'session-1',
|
||||
prompt: 'Write a function',
|
||||
},
|
||||
{
|
||||
ts: Date.now(),
|
||||
worker: 'tcb-alpha',
|
||||
sequence: 2,
|
||||
level: 'info',
|
||||
msg: 'llm.request.finished',
|
||||
span_name: 'llm.request',
|
||||
bead: 'bd-test',
|
||||
session: 'session-1',
|
||||
response: 'Here is a function',
|
||||
model: 'sonnet',
|
||||
tokens: 100,
|
||||
},
|
||||
];
|
||||
|
||||
const sessions = buildConversationSessions(events);
|
||||
expect(sessions).toHaveLength(1);
|
||||
|
||||
const session = sessions[0];
|
||||
expect(session.workerId).toBe('tcb-alpha');
|
||||
expect(session.beadId).toBe('bd-test');
|
||||
expect(session.events).toHaveLength(2);
|
||||
expect(session.events[0].type).toBe('prompt');
|
||||
expect(session.events[1].type).toBe('response');
|
||||
expect(session.totalTokens).toBe(100);
|
||||
});
|
||||
|
||||
it('should build a session with tool calls', () => {
|
||||
const events: LogEvent[] = [
|
||||
{
|
||||
ts: Date.now() - 2000,
|
||||
worker: 'tcb-alpha',
|
||||
sequence: 1,
|
||||
level: 'info',
|
||||
msg: 'tool.call.started',
|
||||
span_name: 'tool.call',
|
||||
bead: 'bd-test',
|
||||
tool: 'Read',
|
||||
tool_args: { file_path: '/test/file.ts' },
|
||||
},
|
||||
{
|
||||
ts: Date.now() - 1000,
|
||||
worker: 'tcb-alpha',
|
||||
sequence: 2,
|
||||
level: 'info',
|
||||
msg: 'tool.call.finished',
|
||||
span_name: 'tool.call',
|
||||
bead: 'bd-test',
|
||||
tool: 'Read',
|
||||
result: 'file contents',
|
||||
},
|
||||
];
|
||||
|
||||
const sessions = buildConversationSessions(events);
|
||||
expect(sessions).toHaveLength(1);
|
||||
|
||||
const session = sessions[0];
|
||||
expect(session.events).toHaveLength(2);
|
||||
expect(session.events[0].type).toBe('tool_call');
|
||||
expect(session.events[1].type).toBe('tool_result');
|
||||
expect(session.toolsUsed).toEqual(['Read']);
|
||||
});
|
||||
|
||||
it('should handle mixed conversation events', () => {
|
||||
const events: LogEvent[] = [
|
||||
{
|
||||
ts: Date.now() - 4000,
|
||||
worker: 'tcb-alpha',
|
||||
sequence: 1,
|
||||
level: 'info',
|
||||
msg: 'llm.request.started',
|
||||
bead: 'bd-test',
|
||||
prompt: 'Read the file',
|
||||
},
|
||||
{
|
||||
ts: Date.now() - 3000,
|
||||
worker: 'tcb-alpha',
|
||||
sequence: 2,
|
||||
level: 'info',
|
||||
msg: 'tool.call.started',
|
||||
bead: 'bd-test',
|
||||
tool: 'Read',
|
||||
args: { file_path: '/test.ts' },
|
||||
},
|
||||
{
|
||||
ts: Date.now() - 2000,
|
||||
worker: 'tcb-alpha',
|
||||
sequence: 3,
|
||||
level: 'info',
|
||||
msg: 'tool.call.finished',
|
||||
bead: 'bd-test',
|
||||
tool: 'Read',
|
||||
result: 'content here',
|
||||
},
|
||||
{
|
||||
ts: Date.now() - 1000,
|
||||
worker: 'tcb-alpha',
|
||||
sequence: 4,
|
||||
level: 'info',
|
||||
msg: 'llm.request.finished',
|
||||
bead: 'bd-test',
|
||||
response: 'The file contains: content here',
|
||||
},
|
||||
];
|
||||
|
||||
const sessions = buildConversationSessions(events);
|
||||
expect(sessions).toHaveLength(1);
|
||||
|
||||
const session = sessions[0];
|
||||
expect(session.events).toHaveLength(4);
|
||||
expect(session.events[0].type).toBe('prompt');
|
||||
expect(session.events[1].type).toBe('tool_call');
|
||||
expect(session.events[2].type).toBe('tool_result');
|
||||
expect(session.events[3].type).toBe('response');
|
||||
expect(session.turnCount).toBe(1);
|
||||
});
|
||||
|
||||
it('should separate sessions by worker and bead', () => {
|
||||
const events: LogEvent[] = [
|
||||
{
|
||||
ts: Date.now(),
|
||||
worker: 'tcb-alpha',
|
||||
sequence: 1,
|
||||
level: 'info',
|
||||
msg: 'llm.request',
|
||||
bead: 'bd-1',
|
||||
prompt: 'Task 1',
|
||||
},
|
||||
{
|
||||
ts: Date.now(),
|
||||
worker: 'tcb-bravo',
|
||||
sequence: 1,
|
||||
level: 'info',
|
||||
msg: 'llm.request',
|
||||
bead: 'bd-2',
|
||||
prompt: 'Task 2',
|
||||
},
|
||||
];
|
||||
|
||||
const sessions = buildConversationSessions(events);
|
||||
expect(sessions).toHaveLength(2);
|
||||
expect(sessions[0].workerId).toBe('tcb-alpha');
|
||||
expect(sessions[1].workerId).toBe('tcb-bravo');
|
||||
});
|
||||
|
||||
it('should return empty array for events with no conversation data', () => {
|
||||
const events: LogEvent[] = [
|
||||
{
|
||||
ts: Date.now(),
|
||||
worker: 'tcb-alpha',
|
||||
sequence: 1,
|
||||
level: 'info',
|
||||
msg: 'worker.state_transition',
|
||||
},
|
||||
{
|
||||
ts: Date.now(),
|
||||
worker: 'tcb-alpha',
|
||||
sequence: 2,
|
||||
level: 'info',
|
||||
msg: 'heartbeat.emitted',
|
||||
},
|
||||
];
|
||||
|
||||
const sessions = buildConversationSessions(events);
|
||||
expect(sessions).toHaveLength(0);
|
||||
});
|
||||
|
||||
it('should extract thinking blocks from assistant events', () => {
|
||||
const events: LogEvent[] = [
|
||||
{
|
||||
ts: Date.now(),
|
||||
worker: 'tcb-alpha',
|
||||
sequence: 1,
|
||||
level: 'info',
|
||||
msg: 'llm.thinking',
|
||||
response: 'Let me think about this...',
|
||||
thinking: true,
|
||||
},
|
||||
];
|
||||
|
||||
const sessions = buildConversationSessions(events);
|
||||
expect(sessions).toHaveLength(1);
|
||||
expect(sessions[0].events[0].type).toBe('thinking');
|
||||
});
|
||||
|
||||
it('should handle tool errors', () => {
|
||||
const events: LogEvent[] = [
|
||||
{
|
||||
ts: Date.now(),
|
||||
worker: 'tcb-alpha',
|
||||
sequence: 1,
|
||||
level: 'info',
|
||||
msg: 'tool.call.finished',
|
||||
tool: 'Read',
|
||||
result: '',
|
||||
error: 'File not found',
|
||||
},
|
||||
];
|
||||
|
||||
const sessions = buildConversationSessions(events);
|
||||
expect(sessions).toHaveLength(1);
|
||||
|
||||
const session = sessions[0];
|
||||
expect(session.events[0].type).toBe('tool_result');
|
||||
const toolResult = session.events[0];
|
||||
if (toolResult.type === 'tool_result') {
|
||||
expect(toolResult.success).toBe(false);
|
||||
expect(toolResult.error).toBe('File not found');
|
||||
}
|
||||
});
|
||||
});
|
||||
|
||||
describe('getWorkerConversationSessions', () => {
|
||||
it('should filter sessions by worker ID', () => {
|
||||
const events: LogEvent[] = [
|
||||
{
|
||||
ts: Date.now(),
|
||||
worker: 'tcb-alpha',
|
||||
sequence: 1,
|
||||
level: 'info',
|
||||
msg: 'llm.request',
|
||||
prompt: 'Alpha task',
|
||||
},
|
||||
{
|
||||
ts: Date.now(),
|
||||
worker: 'tcb-bravo',
|
||||
sequence: 1,
|
||||
level: 'info',
|
||||
msg: 'llm.request',
|
||||
prompt: 'Bravo task',
|
||||
},
|
||||
];
|
||||
|
||||
const alphaSessions = getWorkerConversationSessions(events, 'tcb-alpha');
|
||||
expect(alphaSessions).toHaveLength(1);
|
||||
expect(alphaSessions[0].workerId).toBe('tcb-alpha');
|
||||
});
|
||||
});
|
||||
|
||||
describe('getBeadConversationSession', () => {
|
||||
it('should get session for a specific bead', () => {
|
||||
const events: LogEvent[] = [
|
||||
{
|
||||
ts: Date.now(),
|
||||
worker: 'tcb-alpha',
|
||||
sequence: 1,
|
||||
level: 'info',
|
||||
msg: 'llm.request',
|
||||
bead: 'bd-target',
|
||||
prompt: 'Do this task',
|
||||
},
|
||||
];
|
||||
|
||||
const session = getBeadConversationSession(events, 'bd-target');
|
||||
expect(session).not.toBeNull();
|
||||
expect(session?.beadId).toBe('bd-target');
|
||||
});
|
||||
|
||||
it('should return null for non-existent bead', () => {
|
||||
const events: LogEvent[] = [
|
||||
{
|
||||
ts: Date.now(),
|
||||
worker: 'tcb-alpha',
|
||||
sequence: 1,
|
||||
level: 'info',
|
||||
msg: 'llm.request',
|
||||
bead: 'bd-actual',
|
||||
prompt: 'Task',
|
||||
},
|
||||
];
|
||||
|
||||
const session = getBeadConversationSession(events, 'bd-nonexistent');
|
||||
expect(session).toBeNull();
|
||||
});
|
||||
});
|
||||
|
||||
describe('extractConversationEvents', () => {
|
||||
it('should extract all conversation events from mixed log', () => {
|
||||
const events: LogEvent[] = [
|
||||
{
|
||||
ts: Date.now() - 3000,
|
||||
worker: 'tcb-alpha',
|
||||
sequence: 1,
|
||||
level: 'info',
|
||||
msg: 'worker.state_transition',
|
||||
},
|
||||
{
|
||||
ts: Date.now() - 2000,
|
||||
worker: 'tcb-alpha',
|
||||
sequence: 2,
|
||||
level: 'info',
|
||||
msg: 'llm.request',
|
||||
prompt: 'Hello',
|
||||
},
|
||||
{
|
||||
ts: Date.now() - 1000,
|
||||
worker: 'tcb-alpha',
|
||||
sequence: 3,
|
||||
level: 'info',
|
||||
msg: 'llm.response',
|
||||
response: 'Hi there',
|
||||
},
|
||||
{
|
||||
ts: Date.now(),
|
||||
worker: 'tcb-alpha',
|
||||
sequence: 4,
|
||||
level: 'info',
|
||||
msg: 'heartbeat.emitted',
|
||||
},
|
||||
];
|
||||
|
||||
const convEvents = extractConversationEvents(events);
|
||||
expect(convEvents).toHaveLength(2);
|
||||
expect(convEvents[0].type).toBe('prompt');
|
||||
expect(convEvents[1].type).toBe('response');
|
||||
});
|
||||
|
||||
it('should maintain chronological order across sessions', () => {
|
||||
const baseTime = Date.now();
|
||||
const events: LogEvent[] = [
|
||||
{
|
||||
ts: baseTime - 3000,
|
||||
worker: 'tcb-alpha',
|
||||
sequence: 1,
|
||||
level: 'info',
|
||||
msg: 'llm.request',
|
||||
bead: 'bd-1',
|
||||
prompt: 'First',
|
||||
},
|
||||
{
|
||||
ts: baseTime - 2000,
|
||||
worker: 'tcb-bravo',
|
||||
sequence: 1,
|
||||
level: 'info',
|
||||
msg: 'llm.request',
|
||||
bead: 'bd-2',
|
||||
prompt: 'Second',
|
||||
},
|
||||
{
|
||||
ts: baseTime - 1000,
|
||||
worker: 'tcb-alpha',
|
||||
sequence: 2,
|
||||
level: 'info',
|
||||
msg: 'llm.response',
|
||||
bead: 'bd-1',
|
||||
response: 'First response',
|
||||
},
|
||||
];
|
||||
|
||||
const convEvents = extractConversationEvents(events);
|
||||
expect(convEvents).toHaveLength(3);
|
||||
expect(convEvents[0].type).toBe('prompt');
|
||||
expect((convEvents[0] as any).content).toBe('First');
|
||||
expect(convEvents[1].type).toBe('prompt');
|
||||
expect((convEvents[1] as any).content).toBe('Second');
|
||||
expect(convEvents[2].type).toBe('response');
|
||||
});
|
||||
});
|
||||
});
|
||||
566
src/conversationParser.ts
Normal file
566
src/conversationParser.ts
Normal file
|
|
@ -0,0 +1,566 @@
|
|||
/**
|
||||
* FABRIC Conversation Transcript Parser
|
||||
*
|
||||
* Extracts full conversation transcripts from NEEDLE logs.
|
||||
* Parses OTLP span events (llm.request, tool.call, etc.) into structured conversation events.
|
||||
*
|
||||
* Data sources:
|
||||
* - llm.request.started/finished spans contain prompts, responses, thinking blocks
|
||||
* - tool.call.started/finished spans contain tool invocations and results
|
||||
* - bead.lifecycle spans provide session context (bead_id, worker_id)
|
||||
*/
|
||||
|
||||
import {
|
||||
LogEvent,
|
||||
ConversationEvent,
|
||||
ConversationSession,
|
||||
PromptEvent,
|
||||
ResponseEvent,
|
||||
ThinkingEvent,
|
||||
ToolCallEvent,
|
||||
ToolResultEvent,
|
||||
ConversationEventType,
|
||||
ConversationRole,
|
||||
ToolArgValue,
|
||||
} from './types.js';
|
||||
|
||||
// ============================================
|
||||
// Types
|
||||
// ============================================
|
||||
|
||||
/**
|
||||
* Span event name patterns that indicate conversation content
|
||||
*/
|
||||
const CONVERSATION_SPAN_NAMES = [
|
||||
'llm.request',
|
||||
'llm.prompt',
|
||||
'llm.response',
|
||||
'llm.completion',
|
||||
'llm.chat',
|
||||
'tool.call',
|
||||
'tool.invocation',
|
||||
'tool.result',
|
||||
'tool.response',
|
||||
'bead.prompt_built',
|
||||
'bead.agent_started',
|
||||
'bead.agent_completed',
|
||||
] as const;
|
||||
|
||||
/**
|
||||
* Message event types from spans
|
||||
*/
|
||||
type SpanMessageType =
|
||||
| 'user'
|
||||
| 'system'
|
||||
| 'assistant'
|
||||
| 'tool';
|
||||
|
||||
/**
|
||||
* Parsed span message with content and metadata
|
||||
*/
|
||||
interface ParsedSpanMessage {
|
||||
type: SpanMessageType;
|
||||
content: string;
|
||||
timestamp: number;
|
||||
sequence?: number;
|
||||
toolName?: string;
|
||||
toolArgs?: Record<string, unknown>;
|
||||
toolResult?: string;
|
||||
toolError?: string;
|
||||
isThinking?: boolean;
|
||||
tokens?: number;
|
||||
model?: string;
|
||||
}
|
||||
|
||||
// ============================================
|
||||
// Span Message Extraction
|
||||
// ============================================
|
||||
|
||||
/**
|
||||
* Check if a LogEvent is a conversation-related span event
|
||||
*/
|
||||
export function isConversationSpanEvent(event: LogEvent): boolean {
|
||||
const msg = event.msg.toLowerCase();
|
||||
const spanName = String(event.span_name || '').toLowerCase();
|
||||
|
||||
// Check span_name first (most reliable)
|
||||
for (const pattern of CONVERSATION_SPAN_NAMES) {
|
||||
if (spanName.includes(pattern.toLowerCase())) {
|
||||
return true;
|
||||
}
|
||||
}
|
||||
|
||||
// Check message patterns
|
||||
if (
|
||||
msg.includes('llm.request') ||
|
||||
msg.includes('llm.prompt') ||
|
||||
msg.includes('llm.response') ||
|
||||
msg.includes('llm.completion') ||
|
||||
msg.includes('tool.call') ||
|
||||
msg.includes('tool.invocation') ||
|
||||
msg.includes('tool.result') ||
|
||||
msg.includes('bead.prompt_built') ||
|
||||
msg.includes('bead.agent_started') ||
|
||||
msg.includes('bead.agent_completed')
|
||||
) {
|
||||
return true;
|
||||
}
|
||||
|
||||
// Check for conversation attributes
|
||||
if (
|
||||
event.prompt ||
|
||||
event.response ||
|
||||
event.user_message ||
|
||||
event.assistant_message ||
|
||||
event.tool_name ||
|
||||
event.conversation_role
|
||||
) {
|
||||
return true;
|
||||
}
|
||||
|
||||
return false;
|
||||
}
|
||||
|
||||
/**
|
||||
* Extract conversation data from span event attributes
|
||||
*/
|
||||
function extractSpanMessage(event: LogEvent): ParsedSpanMessage | null {
|
||||
// Try direct fields first
|
||||
if (event.prompt || event.user_message) {
|
||||
return {
|
||||
type: 'user',
|
||||
content: (event.prompt || event.user_message) as string,
|
||||
timestamp: event.ts,
|
||||
sequence: event.sequence,
|
||||
};
|
||||
}
|
||||
|
||||
if (event.response || event.assistant_message) {
|
||||
// Check if it's a thinking block
|
||||
const content = (event.response || event.assistant_message) as string;
|
||||
const isThinking = event.thinking === true || event.is_thinking === true;
|
||||
|
||||
return {
|
||||
type: 'assistant',
|
||||
content,
|
||||
timestamp: event.ts,
|
||||
sequence: event.sequence,
|
||||
isThinking,
|
||||
tokens: event.tokens as number | undefined,
|
||||
model: event.model as string | undefined,
|
||||
};
|
||||
}
|
||||
|
||||
// Check for tool call
|
||||
if (event.tool_name || event.tool) {
|
||||
const toolName = (event.tool_name || event.tool) as string;
|
||||
|
||||
// Check if this is a tool call invocation or result
|
||||
if (event.msg.includes('started') || event.tool_args || event.args) {
|
||||
return {
|
||||
type: 'tool',
|
||||
content: `Calling ${toolName}`,
|
||||
timestamp: event.ts,
|
||||
sequence: event.sequence,
|
||||
toolName,
|
||||
toolArgs: (event.tool_args || event.args || event.arguments) as Record<string, unknown> | undefined,
|
||||
};
|
||||
}
|
||||
|
||||
// Tool result
|
||||
if (event.msg.includes('finished') || event.result || event.tool_result) {
|
||||
const content = (event.result || event.tool_result || '') as string;
|
||||
return {
|
||||
type: 'tool',
|
||||
content: content || 'Tool completed',
|
||||
timestamp: event.ts,
|
||||
sequence: event.sequence,
|
||||
toolName,
|
||||
toolResult: content,
|
||||
toolError: event.error as string | undefined,
|
||||
};
|
||||
}
|
||||
}
|
||||
|
||||
// Try to extract from data field (for OTLP span attributes)
|
||||
const data = event.data as Record<string, unknown> | undefined;
|
||||
if (data) {
|
||||
// Check for LLM request/response in attributes
|
||||
if (data.prompt || data.user_message || data['llm.prompt']) {
|
||||
return {
|
||||
type: 'user',
|
||||
content: (data.prompt || data.user_message || data['llm.prompt']) as string,
|
||||
timestamp: event.ts,
|
||||
sequence: event.sequence,
|
||||
};
|
||||
}
|
||||
|
||||
if (data.response || data.assistant_message || data.completion || data['llm.response']) {
|
||||
const content = (data.response || data.assistant_message || data.completion || data['llm.response']) as string;
|
||||
return {
|
||||
type: 'assistant',
|
||||
content,
|
||||
timestamp: event.ts,
|
||||
sequence: event.sequence,
|
||||
isThinking: data.thinking === true || data.is_thinking === true,
|
||||
tokens: data.tokens as number | undefined,
|
||||
model: data.model as string | undefined,
|
||||
};
|
||||
}
|
||||
|
||||
// Check for tool call in attributes
|
||||
if (data.tool_name || data.tool || data['tool.name']) {
|
||||
const toolName = (data.tool_name || data.tool || data['tool.name']) as string;
|
||||
|
||||
if (event.msg.includes('started') || data.tool_args || data.args || data['tool.arguments']) {
|
||||
return {
|
||||
type: 'tool',
|
||||
content: `Calling ${toolName}`,
|
||||
timestamp: event.ts,
|
||||
sequence: event.sequence,
|
||||
toolName,
|
||||
toolArgs: (data.tool_args || data.args || data['tool.arguments']) as Record<string, unknown> | undefined,
|
||||
};
|
||||
}
|
||||
|
||||
if (event.msg.includes('finished') || data.result || data.tool_result || data['tool.result']) {
|
||||
const content = (data.result || data.tool_result || data['tool.result'] || '') as string;
|
||||
return {
|
||||
type: 'tool',
|
||||
content: content || 'Tool completed',
|
||||
timestamp: event.ts,
|
||||
sequence: event.sequence,
|
||||
toolName,
|
||||
toolResult: content,
|
||||
toolError: data.error as string | undefined,
|
||||
};
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
return null;
|
||||
}
|
||||
|
||||
// ============================================
|
||||
// Conversation Event Construction
|
||||
// ============================================
|
||||
|
||||
let eventIdCounter = 0;
|
||||
|
||||
function generateEventId(): string {
|
||||
return `conv-${Date.now()}-${++eventIdCounter}`;
|
||||
}
|
||||
|
||||
/**
|
||||
* Convert a parsed span message to a ConversationEvent
|
||||
*/
|
||||
function parsedMessageToConversationEvent(
|
||||
message: ParsedSpanMessage,
|
||||
worker: string,
|
||||
bead?: string
|
||||
): ConversationEvent | null {
|
||||
const baseEvent = {
|
||||
id: generateEventId(),
|
||||
ts: message.timestamp,
|
||||
worker,
|
||||
bead,
|
||||
sequence: message.sequence ?? 0,
|
||||
};
|
||||
|
||||
switch (message.type) {
|
||||
case 'user': {
|
||||
const promptEvent: PromptEvent = {
|
||||
...baseEvent,
|
||||
type: 'prompt',
|
||||
role: 'user',
|
||||
content: message.content,
|
||||
tokens: message.tokens,
|
||||
};
|
||||
return promptEvent;
|
||||
}
|
||||
|
||||
case 'assistant': {
|
||||
if (message.isThinking) {
|
||||
const thinkingEvent: ThinkingEvent = {
|
||||
...baseEvent,
|
||||
type: 'thinking',
|
||||
role: 'assistant',
|
||||
content: message.content,
|
||||
isTruncated: false,
|
||||
tokens: message.tokens,
|
||||
durationMs: undefined,
|
||||
};
|
||||
return thinkingEvent;
|
||||
}
|
||||
|
||||
const responseEvent: ResponseEvent = {
|
||||
...baseEvent,
|
||||
type: 'response',
|
||||
role: 'assistant',
|
||||
content: message.content,
|
||||
isTruncated: false,
|
||||
model: message.model,
|
||||
tokens: message.tokens,
|
||||
};
|
||||
return responseEvent;
|
||||
}
|
||||
|
||||
case 'tool': {
|
||||
if (message.toolArgs && message.toolName) {
|
||||
const toolCallEvent: ToolCallEvent = {
|
||||
...baseEvent,
|
||||
type: 'tool_call',
|
||||
role: 'assistant',
|
||||
tool: message.toolName,
|
||||
args: message.toolArgs as Record<string, ToolArgValue>,
|
||||
summary: generateToolSummary(message.toolName, message.toolArgs),
|
||||
};
|
||||
return toolCallEvent;
|
||||
}
|
||||
|
||||
if (message.toolName && message.toolResult !== undefined) {
|
||||
const toolResultEvent: ToolResultEvent = {
|
||||
...baseEvent,
|
||||
type: 'tool_result',
|
||||
role: 'tool',
|
||||
tool: message.toolName,
|
||||
content: message.toolResult,
|
||||
success: !message.toolError,
|
||||
error: message.toolError,
|
||||
isTruncated: false,
|
||||
resultSize: message.toolResult.length,
|
||||
};
|
||||
return toolResultEvent;
|
||||
}
|
||||
|
||||
return null;
|
||||
}
|
||||
|
||||
default:
|
||||
return null;
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Generate a human-readable summary for a tool call
|
||||
*/
|
||||
function generateToolSummary(tool: string, args: Record<string, unknown>): string {
|
||||
switch (tool) {
|
||||
case 'Read':
|
||||
return `Read ${args.file_path || args.path || 'file'}`;
|
||||
case 'Edit':
|
||||
return `Edit ${args.file_path || args.path || 'file'}`;
|
||||
case 'Write':
|
||||
return `Write ${args.file_path || args.path || 'file'}`;
|
||||
case 'Bash':
|
||||
return `Run: ${(args.command as string)?.slice(0, 50) || 'command'}`;
|
||||
case 'Grep':
|
||||
return `Search: ${args.pattern || 'pattern'}`;
|
||||
case 'Glob':
|
||||
return `Find: ${args.pattern || 'files'}`;
|
||||
default:
|
||||
return `${tool}()`;
|
||||
}
|
||||
}
|
||||
|
||||
// ============================================
|
||||
// Session Building
|
||||
// ============================================
|
||||
|
||||
/**
|
||||
* Build a ConversationSession from a list of log events
|
||||
*
|
||||
* Groups events by (worker, bead, session) and extracts conversation
|
||||
* data from span events within each session.
|
||||
*/
|
||||
export function buildConversationSessions(
|
||||
events: LogEvent[]
|
||||
): ConversationSession[] {
|
||||
// Group events by potential session keys
|
||||
const sessionGroups = groupEventsBySession(events);
|
||||
const sessions: ConversationSession[] = [];
|
||||
|
||||
for (const [sessionKey, sessionEvents] of sessionGroups) {
|
||||
const session = buildSessionFromEvents(sessionKey, sessionEvents);
|
||||
if (session && session.events.length > 0) {
|
||||
sessions.push(session);
|
||||
}
|
||||
}
|
||||
|
||||
return sessions;
|
||||
}
|
||||
|
||||
/**
|
||||
* Session key for grouping events
|
||||
*/
|
||||
interface SessionKey {
|
||||
worker: string;
|
||||
bead?: string;
|
||||
session?: string;
|
||||
}
|
||||
|
||||
function sessionKeyToString(key: SessionKey): string {
|
||||
return `${key.worker}:${key.bead || 'none'}:${key.session || 'none'}`;
|
||||
}
|
||||
|
||||
/**
|
||||
* Group events by (worker, bead, session)
|
||||
*/
|
||||
function groupEventsBySession(events: LogEvent[]): Map<string, LogEvent[]> {
|
||||
const groups = new Map<string, LogEvent[]>();
|
||||
|
||||
for (const event of events) {
|
||||
// Skip events that aren't conversation-related
|
||||
if (!isConversationSpanEvent(event)) {
|
||||
continue;
|
||||
}
|
||||
|
||||
const key: SessionKey = {
|
||||
worker: event.worker,
|
||||
bead: event.bead,
|
||||
session: event.session,
|
||||
};
|
||||
|
||||
const keyStr = sessionKeyToString(key);
|
||||
const group = groups.get(keyStr) || [];
|
||||
group.push(event);
|
||||
groups.set(keyStr, group);
|
||||
}
|
||||
|
||||
return groups;
|
||||
}
|
||||
|
||||
/**
|
||||
* Build a single ConversationSession from grouped events
|
||||
*/
|
||||
function buildSessionFromEvents(
|
||||
sessionKey: string,
|
||||
events: LogEvent[]
|
||||
): ConversationSession | null {
|
||||
if (events.length === 0) return null;
|
||||
|
||||
// Sort by sequence or timestamp
|
||||
const sorted = [...events].sort((a, b) => {
|
||||
const seqA = a.sequence ?? a.ts;
|
||||
const seqB = b.sequence ?? b.ts;
|
||||
return seqA - seqB;
|
||||
});
|
||||
|
||||
// Extract conversation events
|
||||
const conversationEvents: ConversationEvent[] = [];
|
||||
let sequence = 0;
|
||||
|
||||
for (const event of sorted) {
|
||||
const message = extractSpanMessage(event);
|
||||
if (!message) continue;
|
||||
|
||||
message.sequence = sequence++;
|
||||
|
||||
const convEvent = parsedMessageToConversationEvent(
|
||||
message,
|
||||
event.worker,
|
||||
event.bead
|
||||
);
|
||||
|
||||
if (convEvent) {
|
||||
conversationEvents.push(convEvent);
|
||||
}
|
||||
}
|
||||
|
||||
if (conversationEvents.length === 0) {
|
||||
return null;
|
||||
}
|
||||
|
||||
// Build session metadata
|
||||
const firstEvent = sorted[0];
|
||||
const lastEvent = sorted[sorted.length - 1];
|
||||
|
||||
// Extract worker, bead, session from first event
|
||||
const workerId = firstEvent.worker;
|
||||
const beadId = firstEvent.bead;
|
||||
const sessionId = firstEvent.session || `${workerId}-${firstEvent.ts}`;
|
||||
|
||||
// Calculate tokens
|
||||
const totalTokens = conversationEvents.reduce((sum, e) => sum + (e.tokens || 0), 0);
|
||||
|
||||
// Extract tools used
|
||||
const toolsUsed = [
|
||||
...new Set(
|
||||
conversationEvents
|
||||
.filter((e): e is ToolCallEvent => e.type === 'tool_call')
|
||||
.map((e) => e.tool)
|
||||
),
|
||||
];
|
||||
|
||||
// Count turns (prompt + response pairs)
|
||||
let turnCount = 0;
|
||||
let inTurn = false;
|
||||
for (const event of conversationEvents) {
|
||||
if (event.type === 'prompt') {
|
||||
inTurn = true;
|
||||
} else if (event.type === 'response' && inTurn) {
|
||||
turnCount++;
|
||||
inTurn = false;
|
||||
}
|
||||
}
|
||||
|
||||
return {
|
||||
id: sessionId,
|
||||
workerId,
|
||||
beadId,
|
||||
startTime: firstEvent.ts,
|
||||
endTime: lastEvent.ts,
|
||||
events: conversationEvents,
|
||||
totalTokens,
|
||||
turnCount,
|
||||
toolsUsed,
|
||||
isActive: false, // Sessions from historical logs are complete
|
||||
};
|
||||
}
|
||||
|
||||
/**
|
||||
* Get conversation sessions for a specific worker
|
||||
*/
|
||||
export function getWorkerConversationSessions(
|
||||
events: LogEvent[],
|
||||
workerId: string
|
||||
): ConversationSession[] {
|
||||
const workerEvents = events.filter(e => e.worker === workerId);
|
||||
return buildConversationSessions(workerEvents);
|
||||
}
|
||||
|
||||
/**
|
||||
* Get conversation session for a specific bead
|
||||
*/
|
||||
export function getBeadConversationSession(
|
||||
events: LogEvent[],
|
||||
beadId: string
|
||||
): ConversationSession | null {
|
||||
const beadEvents = events.filter(e => e.bead === beadId);
|
||||
const sessions = buildConversationSessions(beadEvents);
|
||||
return sessions[0] || null;
|
||||
}
|
||||
|
||||
/**
|
||||
* Extract conversation events from a list of log events
|
||||
*
|
||||
* This is the main entry point for getting conversation data
|
||||
* from raw log events.
|
||||
*/
|
||||
export function extractConversationEvents(
|
||||
events: LogEvent[]
|
||||
): ConversationEvent[] {
|
||||
const sessions = buildConversationSessions(events);
|
||||
const allEvents: ConversationEvent[] = [];
|
||||
|
||||
for (const session of sessions) {
|
||||
allEvents.push(...session.events);
|
||||
}
|
||||
|
||||
// Sort all events by timestamp/sequence
|
||||
return allEvents.sort((a, b) => {
|
||||
const seqA = a.sequence ?? a.ts;
|
||||
const seqB = b.sequence ?? b.ts;
|
||||
return seqA - seqB;
|
||||
});
|
||||
}
|
||||
Loading…
Add table
Reference in a new issue