agent-ecosystem/src/main/services/team/TeamMemberLogsFinder.ts
iliya 85fa86f1be refactor: optimize TeamMemberLogsFinder for efficient subagent file processing
- Introduced a new method to collect subagent file candidates, streamlining the file scanning process.
- Implemented parallel processing of subagent files with a concurrency limit to enhance performance.
- Improved error handling during file parsing, ensuring that failures do not disrupt the overall scanning process.
- Added unit tests to verify that member logs are returned sorted by start time in descending order.
2026-03-14 19:44:28 +02:00

1408 lines
48 KiB
TypeScript

import { encodePath, extractBaseDir, getProjectsBasePath } from '@main/utils/pathDecoder';
import { createLogger } from '@shared/utils/logger';
import { parseAllTeammateMessages } from '@shared/utils/teammateMessageParser';
import { createReadStream } from 'fs';
import * as fs from 'fs/promises';
import * as path from 'path';
import * as readline from 'readline';
import { TeamConfigReader } from './TeamConfigReader';
import { TeamInboxReader } from './TeamInboxReader';
import { TeamMembersMetaStore } from './TeamMembersMetaStore';
import type { MemberLogSummary, MemberSubagentLogSummary } from '@shared/types';
const logger = createLogger('Service:TeamMemberLogsFinder');
/**
* Phase 1: How many lines to scan for member attribution.
* Detection signals (process.team.memberName, "You are {name}", routing.sender)
* appear in the first ~10 lines, so 50 is very conservative.
*/
const ATTRIBUTION_SCAN_LINES = 50;
/** Grace before task creation — logs cannot reference a task before it exists. */
const TASK_SINCE_GRACE_MS = 2 * 60 * 1000;
const FILE_MENTIONS_CACHE_MAX = 1000;
/** Max concurrent file reads during parallel scan phases. */
const SCAN_CONCURRENCY = 15;
/** Signal sources for subagent member attribution, ordered by reliability. */
type AttributionSignalSource = 'process_team' | 'routing_sender' | 'teammate_id' | 'text_mention';
interface DetectionSignal {
member: string;
source: AttributionSignalSource;
}
/**
* Precedence order for attribution signals (most reliable first).
* - process_team: from system init message — written by CLI, definitive
* - routing_sender: from toolUseResult.routing — identifies the actual agent
* - teammate_id: from <teammate-message> XML — identifies the message SENDER, not the agent
* - text_mention: regex match of member name in text — lowest reliability
*/
const SIGNAL_PRECEDENCE: readonly AttributionSignalSource[] = [
'process_team',
'routing_sender',
'teammate_id',
'text_mention',
];
interface StreamedMetadata {
firstTimestamp: string | null;
lastTimestamp: string | null;
messageCount: number;
}
/** Result of attributing a subagent file to a team member. */
interface SubagentAttribution {
detectedMember: string;
description: string;
firstTimestamp: string | null;
}
function trimTrailingSlashes(value: string): string {
let end = value.length;
while (end > 0) {
const ch = value.charCodeAt(end - 1);
// '/' or '\'
if (ch === 47 || ch === 92) {
end--;
continue;
}
break;
}
return end === value.length ? value : value.slice(0, end);
}
export class TeamMemberLogsFinder {
private readonly fileMentionsCache = new Map<string, boolean>();
constructor(
private readonly configReader: TeamConfigReader = new TeamConfigReader(),
private readonly inboxReader: TeamInboxReader = new TeamInboxReader(),
private readonly membersMetaStore: TeamMembersMetaStore = new TeamMembersMetaStore()
) {}
async findMemberLogs(teamName: string, memberName: string): Promise<MemberLogSummary[]> {
const discovery = await this.discoverMemberFiles(teamName, memberName);
if (!discovery) return [];
const { projectDir, projectId, config, sessionIds, knownMembers, isLeadMember } = discovery;
const results: MemberLogSummary[] = [];
const leadMemberName =
config.members?.find((m) => m?.agentType === 'team-lead')?.name?.trim() || 'team-lead';
if (isLeadMember && config.leadSessionId) {
const leadJsonl = path.join(projectDir, `${config.leadSessionId}.jsonl`);
const leadSummary = await this.parseLeadSessionSummary(
leadJsonl,
projectId,
config.leadSessionId,
leadMemberName
);
if (leadSummary) {
results.push(leadSummary);
}
}
// ── Collect and parallel-scan subagent files ──
const candidates = await this.collectSubagentCandidates(projectDir, sessionIds);
const settled: (MemberSubagentLogSummary | null)[] = new Array(candidates.length).fill(null);
let nextIdx = 0;
const scanWorker = async (): Promise<void> => {
while (nextIdx < candidates.length) {
const idx = nextIdx++;
const c = candidates[idx];
try {
const summary = await this.parseSubagentSummary(
c.filePath,
projectId,
c.sessionId,
c.fileName,
memberName,
knownMembers
);
if (summary) settled[idx] = summary;
} catch (err) {
logger.warn(`Failed to parse subagent summary: ${c.filePath}`, err);
}
}
};
await Promise.all(
Array.from({ length: Math.min(SCAN_CONCURRENCY, candidates.length) }, () => scanWorker())
);
for (const s of settled) {
if (s) results.push(s);
}
return results.sort(
(a, b) => new Date(b.startTime).getTime() - new Date(a.startTime).getTime()
);
}
/**
* Returns session logs that reference the given task (TaskCreate, TaskUpdate, comments, etc.).
* When the task is in_progress and has an owner, also includes that owner's session logs so
* the executor's current activity is visible even before the JSONL mentions the task id.
*/
async findLogsForTask(
teamName: string,
taskId: string,
options?: {
owner?: string;
status?: string;
intervals?: { startedAt: string; completedAt?: string }[];
since?: string;
}
): Promise<MemberLogSummary[]> {
const t0 = performance.now();
const discovery = await this.discoverProjectSessions(teamName);
const tDiscovery = performance.now();
if (!discovery) {
console.log(
`[perf] findLogsForTask(${taskId}) discovery=null ${(tDiscovery - t0).toFixed(0)}ms`
);
return [];
}
const sinceMs = this.deriveSinceMs(options);
const { projectDir, projectId, config, sessionIds, knownMembers } = discovery;
const results: MemberLogSummary[] = [];
const leadMemberName =
config.members?.find((m) => m?.agentType === 'team-lead')?.name?.trim() || 'team-lead';
if (config.leadSessionId) {
const leadJsonl = path.join(projectDir, `${config.leadSessionId}.jsonl`);
try {
await fs.access(leadJsonl);
if (await this.fileMentionsTaskIdCached(leadJsonl, teamName, taskId, true, sinceMs)) {
const leadSummary = await this.parseLeadSessionSummary(
leadJsonl,
projectId,
config.leadSessionId,
leadMemberName
);
if (leadSummary) results.push(leadSummary);
}
} catch {
// file missing or unreadable
}
}
const tLead = performance.now();
// ── Collect all subagent file candidates ──
const candidates = await this.collectSubagentCandidates(projectDir, sessionIds);
// ── Parallel scan with concurrency limit ──
const settled: (MemberLogSummary | null)[] = new Array(candidates.length).fill(null);
let nextIdx = 0;
let mentionHits = 0;
const scanWorker = async (): Promise<void> => {
while (nextIdx < candidates.length) {
const idx = nextIdx++;
const c = candidates[idx];
try {
if (!(await this.fileMentionsTaskIdCached(c.filePath, teamName, taskId, false, sinceMs)))
continue;
mentionHits++;
const attribution = await this.attributeSubagent(c.filePath, knownMembers);
if (!attribution) continue;
const summary = await this.parseSubagentSummary(
c.filePath,
projectId,
c.sessionId,
c.fileName,
attribution.detectedMember,
knownMembers,
attribution
);
if (summary) settled[idx] = summary;
} catch (err) {
logger.warn(`Failed to scan subagent file: ${c.filePath}`, err);
}
}
};
await Promise.all(
Array.from({ length: Math.min(SCAN_CONCURRENCY, candidates.length) }, () => scanWorker())
);
for (const s of settled) {
if (s) results.push(s);
}
const totalFiles = candidates.length;
const tScan = performance.now();
const normalizedOwner =
typeof options?.owner === 'string' ? options.owner.trim() : options?.owner;
const isLeadOwner =
typeof normalizedOwner === 'string' &&
normalizedOwner.length > 0 &&
normalizedOwner.toLowerCase() === leadMemberName.toLowerCase();
const ownerRelevantStatus =
options?.status === 'in_progress' || options?.status === 'completed';
const includeOwnerSessions =
ownerRelevantStatus &&
typeof normalizedOwner === 'string' &&
normalizedOwner.length > 0 &&
!isLeadOwner;
if (includeOwnerSessions) {
const ownerLogs = await this.findMemberLogs(teamName, normalizedOwner);
const TASK_LOG_INTERVAL_GRACE_MS = 10_000;
const fallbackRecentMs = 30 * 60_000; // if caller doesn't supply intervals/since, avoid pulling in old owner history
const now = Date.now();
const normalizedIntervals = Array.isArray(options?.intervals)
? options.intervals
.map((i) => {
const startMs = Date.parse(i.startedAt);
const endMsRaw =
typeof i.completedAt === 'string' ? Date.parse(i.completedAt) : Number.NaN;
const endMs = Number.isFinite(endMsRaw) ? endMsRaw : null;
return Number.isFinite(startMs) ? { startMs, endMs } : null;
})
.filter((v): v is { startMs: number; endMs: number | null } => v !== null)
: [];
// Back-compat: single since timestamp -> treat as open interval.
const sinceMsRaw = typeof options?.since === 'string' ? Date.parse(options.since) : NaN;
const sinceStartMs = Number.isFinite(sinceMsRaw) ? sinceMsRaw : null;
const effectiveIntervals =
normalizedIntervals.length > 0
? normalizedIntervals
: sinceStartMs != null
? [{ startMs: sinceStartMs, endMs: null }]
: [];
const filteredOwnerLogs = ownerLogs.filter((log) => {
if (log.isOngoing) return true;
const startMs = new Date(log.startTime).getTime();
if (!Number.isFinite(startMs)) return false;
const durationMs =
typeof log.durationMs === 'number' && log.durationMs > 0 ? log.durationMs : 0;
const endMs = startMs + durationMs;
if (effectiveIntervals.length > 0) {
return this.logOverlapsIntervals(
startMs,
endMs,
effectiveIntervals,
now,
TASK_LOG_INTERVAL_GRACE_MS
);
}
return startMs >= now - fallbackRecentMs;
});
const seen = new Set<string>();
for (const log of results) {
const key =
log.kind === 'subagent'
? `subagent:${log.sessionId}:${log.subagentId}`
: `lead:${log.sessionId}`;
seen.add(key);
}
for (const log of filteredOwnerLogs) {
const key =
log.kind === 'subagent'
? `subagent:${log.sessionId}:${log.subagentId}`
: `lead:${log.sessionId}`;
if (!seen.has(key)) {
seen.add(key);
results.push(log);
}
}
}
const tOwner = performance.now();
const sorted = results.sort(
(a, b) => new Date(b.startTime).getTime() - new Date(a.startTime).getTime()
);
const tTotal = performance.now();
console.log(
`[perf] findLogsForTask(${taskId}@${teamName}) ` +
`total=${(tTotal - t0).toFixed(0)}ms | ` +
`discovery=${(tDiscovery - t0).toFixed(0)}ms | ` +
`lead=${(tLead - tDiscovery).toFixed(0)}ms | ` +
`scan=${(tScan - tLead).toFixed(0)}ms (${totalFiles} files, ${mentionHits} hits) | ` +
`owner=${(tOwner - tScan).toFixed(0)}ms | ` +
`sessions=${sessionIds.length} | results=${sorted.length}`
);
return sorted;
}
/**
* Fast path for change extraction: returns task-related JSONL file refs directly without
* building full MemberLogSummary metadata for every matched log.
*/
async findLogFileRefsForTask(
teamName: string,
taskId: string,
options?: {
owner?: string;
status?: string;
intervals?: { startedAt: string; completedAt?: string }[];
since?: string;
}
): Promise<{ filePath: string; memberName: string }[]> {
const t0 = performance.now();
const discovery = await this.discoverProjectSessions(teamName);
const tDiscovery = performance.now();
if (!discovery) {
console.log(
`[perf] findLogFileRefsForTask(${taskId}) discovery=null ${(tDiscovery - t0).toFixed(0)}ms`
);
return [];
}
const sinceMs = this.deriveSinceMs(options);
const { projectDir, config, sessionIds, knownMembers } = discovery;
const refs: { filePath: string; memberName: string; sortTime: number }[] = [];
const seen = new Set<string>();
const leadMemberName =
config.members?.find((m) => m?.agentType === 'team-lead')?.name?.trim() || 'team-lead';
const pushRef = (filePath: string, memberName: string, sortTime = 0): void => {
const key = `${memberName.toLowerCase()}:${filePath}`;
if (seen.has(key)) return;
seen.add(key);
refs.push({ filePath, memberName, sortTime });
};
if (config.leadSessionId) {
const leadJsonl = path.join(projectDir, `${config.leadSessionId}.jsonl`);
try {
await fs.access(leadJsonl);
if (await this.fileMentionsTaskIdCached(leadJsonl, teamName, taskId, true, sinceMs)) {
const firstTimestamp = await this.probeFirstTimestamp(leadJsonl);
pushRef(leadJsonl, leadMemberName, await this.getSortTime(leadJsonl, firstTimestamp));
}
} catch {
// file missing or unreadable
}
}
const tLead = performance.now();
// ── Collect all subagent file candidates ──
const candidates = await this.collectSubagentCandidates(projectDir, sessionIds);
// ── Parallel scan with concurrency limit ──
let nextIdx = 0;
let mentionHits = 0;
const scanWorker = async (): Promise<void> => {
while (nextIdx < candidates.length) {
const idx = nextIdx++;
const c = candidates[idx];
try {
if (!(await this.fileMentionsTaskIdCached(c.filePath, teamName, taskId, false, sinceMs)))
continue;
mentionHits++;
const attribution = await this.attributeSubagent(c.filePath, knownMembers);
if (!attribution) continue;
pushRef(
c.filePath,
attribution.detectedMember,
await this.getSortTime(c.filePath, attribution.firstTimestamp)
);
} catch (err) {
logger.warn(`Failed to scan subagent file: ${c.filePath}`, err);
}
}
};
await Promise.all(
Array.from({ length: Math.min(SCAN_CONCURRENCY, candidates.length) }, () => scanWorker())
);
const totalFiles = candidates.length;
const tScan = performance.now();
const normalizedOwner =
typeof options?.owner === 'string' ? options.owner.trim() : options?.owner;
const isLeadOwner =
typeof normalizedOwner === 'string' &&
normalizedOwner.length > 0 &&
normalizedOwner.toLowerCase() === leadMemberName.toLowerCase();
const ownerRelevantStatus =
options?.status === 'in_progress' || options?.status === 'completed';
const includeOwnerSessions =
ownerRelevantStatus &&
typeof normalizedOwner === 'string' &&
normalizedOwner.length > 0 &&
!isLeadOwner;
if (includeOwnerSessions) {
const ownerLogs = await this.findMemberLogs(teamName, normalizedOwner);
const TASK_LOG_INTERVAL_GRACE_MS = 10_000;
const fallbackRecentMs = 30 * 60_000;
const now = Date.now();
const normalizedIntervals = Array.isArray(options?.intervals)
? options.intervals
.map((i) => {
const startMs = Date.parse(i.startedAt);
const endMsRaw =
typeof i.completedAt === 'string' ? Date.parse(i.completedAt) : Number.NaN;
const endMs = Number.isFinite(endMsRaw) ? endMsRaw : null;
return Number.isFinite(startMs) ? { startMs, endMs } : null;
})
.filter((v): v is { startMs: number; endMs: number | null } => v !== null)
: [];
const sinceMsRaw = typeof options?.since === 'string' ? Date.parse(options.since) : NaN;
const sinceStartMs = Number.isFinite(sinceMsRaw) ? sinceMsRaw : null;
const effectiveIntervals =
normalizedIntervals.length > 0
? normalizedIntervals
: sinceStartMs != null
? [{ startMs: sinceStartMs, endMs: null }]
: [];
for (const log of ownerLogs) {
if (!log.filePath) continue;
if (!log.isOngoing) {
const startMs = new Date(log.startTime).getTime();
if (!Number.isFinite(startMs)) continue;
const durationMs =
typeof log.durationMs === 'number' && log.durationMs > 0 ? log.durationMs : 0;
const endMs = startMs + durationMs;
if (effectiveIntervals.length > 0) {
if (
!this.logOverlapsIntervals(
startMs,
endMs,
effectiveIntervals,
now,
TASK_LOG_INTERVAL_GRACE_MS
)
) {
continue;
}
} else if (startMs < now - fallbackRecentMs) {
continue;
}
}
pushRef(
log.filePath,
log.memberName ?? normalizedOwner,
Number.isFinite(new Date(log.startTime).getTime()) ? new Date(log.startTime).getTime() : 0
);
}
}
const tOwner = performance.now();
const sortedRefs = [...refs].sort((a, b) => b.sortTime - a.sortTime);
const tTotal = performance.now();
console.log(
`[perf] findLogFileRefsForTask(${taskId}@${teamName}) ` +
`total=${(tTotal - t0).toFixed(0)}ms | ` +
`discovery=${(tDiscovery - t0).toFixed(0)}ms | ` +
`lead=${(tLead - tDiscovery).toFixed(0)}ms | ` +
`scan=${(tScan - tLead).toFixed(0)}ms (${totalFiles} files, ${mentionHits} hits) | ` +
`owner=${(tOwner - tScan).toFixed(0)}ms | ` +
`sessions=${sessionIds.length} | results=${sortedRefs.length}`
);
return sortedRefs.map(({ filePath, memberName }) => ({ filePath, memberName }));
}
/**
* Returns absolute paths to all JSONL files belonging to the specified member.
* Uses the same discovery logic as findMemberLogs but collects file paths.
*/
async findMemberLogPaths(teamName: string, memberName: string): Promise<string[]> {
const discovery = await this.discoverMemberFiles(teamName, memberName);
if (!discovery) return [];
const { projectDir, config, sessionIds, knownMembers, isLeadMember } = discovery;
const paths: string[] = [];
if (isLeadMember && config.leadSessionId) {
const leadJsonl = path.join(projectDir, `${config.leadSessionId}.jsonl`);
try {
await fs.access(leadJsonl);
paths.push(leadJsonl);
} catch {
// File doesn't exist
}
}
for (const sessionId of sessionIds) {
const subagentsDir = path.join(projectDir, sessionId, 'subagents');
let files: string[];
try {
files = await fs.readdir(subagentsDir);
} catch {
continue;
}
for (const file of files) {
if (!file.startsWith('agent-') || !file.endsWith('.jsonl')) continue;
if (file.startsWith('agent-acompact')) continue;
const filePath = path.join(subagentsDir, file);
// Quick attribution check — only Phase 1 (no full-file streaming)
const attribution = await this.attributeSubagent(filePath, knownMembers);
if (attribution?.detectedMember.toLowerCase() === memberName.trim().toLowerCase()) {
paths.push(filePath);
}
}
}
return paths;
}
/**
* Fast marker probe for task-related logs.
* Prefer structured MCP/TaskUpdate markers for modern sessions.
*/
async hasTaskUpdateMarker(filePath: string, taskId: string): Promise<boolean> {
const stream = createReadStream(filePath, { encoding: 'utf8' });
const rl = readline.createInterface({ input: stream, crlfDelay: Infinity });
const escapedTaskId = taskId.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
const pattern = new RegExp(`"taskId"\\s*:\\s*"${escapedTaskId}"`);
try {
for await (const line of rl) {
if (line.includes('TaskUpdate') && pattern.test(line)) {
rl.close();
stream.destroy();
return true;
}
if (
(line.includes('"task_start"') ||
line.includes('"task_complete"') ||
line.includes('"task_set_status"')) &&
pattern.test(line)
) {
rl.close();
stream.destroy();
return true;
}
}
} catch {
// ignore read errors
}
rl.close();
stream.destroy();
return false;
}
private async discoverProjectSessions(teamName: string): Promise<{
projectDir: string;
projectId: string;
config: NonNullable<Awaited<ReturnType<TeamConfigReader['getConfig']>>>;
sessionIds: string[];
knownMembers: Set<string>;
} | null> {
const config = await this.configReader.getConfig(teamName);
if (!config?.projectPath) {
logger.debug(`No projectPath for team "${teamName}"`);
return null;
}
const normalizedProjectPath = trimTrailingSlashes(config.projectPath);
let projectId = encodePath(normalizedProjectPath);
let baseDir = extractBaseDir(projectId);
let projectDir = path.join(getProjectsBasePath(), baseDir);
// If the encoded directory doesn't exist (symlink/cwd mismatch), fall back to locating
// the project directory by leadSessionId which is unique and reliable.
try {
const stat = await fs.stat(projectDir);
if (!stat.isDirectory()) {
throw new Error('not a directory');
}
} catch {
const leadSessionId =
typeof config.leadSessionId === 'string' && config.leadSessionId.trim().length > 0
? config.leadSessionId.trim()
: null;
if (leadSessionId) {
const projectsBase = getProjectsBasePath();
try {
const entries = await fs.readdir(projectsBase, { withFileTypes: true });
for (const entry of entries) {
if (!entry.isDirectory()) continue;
const candidateDir = path.join(projectsBase, entry.name);
const leadPath = path.join(candidateDir, `${leadSessionId}.jsonl`);
try {
await fs.access(leadPath);
projectDir = candidateDir;
projectId = entry.name;
baseDir = entry.name;
break;
} catch {
// not this project
}
}
} catch {
// ignore
}
}
}
const knownSessionIds = new Set<string>();
if (config.leadSessionId) {
knownSessionIds.add(config.leadSessionId);
}
if (Array.isArray(config.sessionHistory)) {
for (const sid of config.sessionHistory) {
if (typeof sid === 'string' && sid.trim().length > 0) {
knownSessionIds.add(sid.trim());
}
}
}
const discoveredSessionIds = await this.listSessionDirs(projectDir);
let sessionIds: string[];
if (knownSessionIds.size > 0) {
const verified: string[] = [];
for (const sid of knownSessionIds) {
const sidDir = path.join(projectDir, sid);
try {
const stat = await fs.stat(sidDir);
if (stat.isDirectory()) verified.push(sid);
} catch {
// dir doesn't exist
}
}
// Prefer config-backed sessions first, but also include any live session dirs that have
// appeared on disk and are not yet reflected in config/sessionHistory.
sessionIds = Array.from(new Set([...verified, ...discoveredSessionIds]));
} else {
sessionIds = discoveredSessionIds;
}
const knownMembers = new Set<string>(
(config.members ?? [])
.map((member) => member.name?.trim().toLowerCase())
.filter((name): name is string => Boolean(name && name.length > 0))
);
try {
const metaMembers = await this.membersMetaStore.getMembers(teamName);
for (const member of metaMembers) {
const normalized = member.name.trim().toLowerCase();
if (normalized.length > 0) knownMembers.add(normalized);
}
} catch {
// best-effort
}
try {
const inboxMembers = await this.inboxReader.listInboxNames(teamName);
for (const name of inboxMembers) {
const normalized = name.trim().toLowerCase();
if (normalized.length > 0) knownMembers.add(normalized);
}
} catch {
// best-effort
}
return { projectDir, projectId, config, sessionIds, knownMembers };
}
private async discoverMemberFiles(
teamName: string,
memberName: string
): Promise<{
projectDir: string;
projectId: string;
config: NonNullable<Awaited<ReturnType<TeamConfigReader['getConfig']>>>;
sessionIds: string[];
knownMembers: Set<string>;
isLeadMember: boolean;
} | null> {
const discovery = await this.discoverProjectSessions(teamName);
if (!discovery) return null;
const { config } = discovery;
const leadMemberName =
config.members?.find((m) => m?.agentType === 'team-lead')?.name?.trim() || 'team-lead';
const isLeadMember = leadMemberName.toLowerCase() === memberName.trim().toLowerCase();
return { ...discovery, isLeadMember };
}
/**
* Collect all subagent JSONL file candidates across session directories.
* Filters out non-agent files and compact files (agent-acompact*).
*/
private async collectSubagentCandidates(
projectDir: string,
sessionIds: string[]
): Promise<{ filePath: string; sessionId: string; fileName: string }[]> {
const candidates: { filePath: string; sessionId: string; fileName: string }[] = [];
for (const sessionId of sessionIds) {
const subagentsDir = path.join(projectDir, sessionId, 'subagents');
let dirFiles: string[];
try {
dirFiles = await fs.readdir(subagentsDir);
} catch {
continue;
}
for (const f of dirFiles) {
if (!f.startsWith('agent-') || !f.endsWith('.jsonl') || f.startsWith('agent-acompact'))
continue;
candidates.push({ filePath: path.join(subagentsDir, f), sessionId, fileName: f });
}
}
return candidates;
}
private deriveSinceMs(options?: {
intervals?: { startedAt: string; completedAt?: string }[];
since?: string;
}): number | null {
const sinceRaw = typeof options?.since === 'string' ? options.since : null;
if (sinceRaw) {
const ms = Date.parse(sinceRaw);
return Number.isFinite(ms) ? ms : null;
}
const intervals = options?.intervals;
if (!Array.isArray(intervals) || intervals.length === 0) return null;
let earliest = Number.POSITIVE_INFINITY;
for (const i of intervals) {
if (typeof i.startedAt === 'string') {
const ms = Date.parse(i.startedAt);
if (Number.isFinite(ms) && ms < earliest) earliest = ms;
}
}
if (!Number.isFinite(earliest) || earliest === Number.POSITIVE_INFINITY) return null;
return earliest - TASK_SINCE_GRACE_MS;
}
private logOverlapsIntervals(
logStartMs: number,
logEndMs: number,
intervals: { startMs: number; endMs: number | null }[],
now: number,
graceMs: number
): boolean {
for (const it of intervals) {
const start = it.startMs - graceMs;
const end = (it.endMs ?? now) + graceMs;
if (logStartMs <= end && logEndMs >= start) return true;
}
return false;
}
private async fileMentionsTaskIdCached(
filePath: string,
teamName: string,
taskId: string,
assumeTeam: boolean,
sinceMs: number | null
): Promise<boolean> {
let mtimeMs: number;
try {
const stat = await fs.stat(filePath);
mtimeMs = stat.mtimeMs;
} catch {
return false;
}
if (sinceMs != null && mtimeMs < sinceMs - TASK_SINCE_GRACE_MS) {
return false;
}
const cacheKey = `${filePath}:${mtimeMs}:${taskId}:${teamName}:${assumeTeam}`;
const cached = this.fileMentionsCache.get(cacheKey);
if (cached !== undefined) return cached;
const result = await this.fileMentionsTaskId(filePath, teamName, taskId, assumeTeam);
this.fileMentionsCache.set(cacheKey, result);
if (this.fileMentionsCache.size > FILE_MENTIONS_CACHE_MAX) {
const keys = [...this.fileMentionsCache.keys()];
for (let i = 0; i < Math.min(keys.length / 2, 50); i++) {
this.fileMentionsCache.delete(keys[i]);
}
}
return result;
}
private async fileMentionsTaskId(
filePath: string,
teamName: string,
taskId: string,
assumeTeam: boolean = false
): Promise<boolean> {
const teamLower = teamName.trim().toLowerCase();
const taskIdStr = taskId.trim();
const extractTaskIdFromUnknown = (raw: unknown): string | null => {
if (typeof raw === 'string') return raw.trim();
if (typeof raw === 'number' && Number.isFinite(raw)) return String(raw);
return null;
};
const extractTeamFromInput = (input: Record<string, unknown>): string | null => {
const raw =
typeof input.team_name === 'string'
? input.team_name
: typeof input.teamName === 'string'
? input.teamName
: typeof input.team === 'string'
? input.team
: null;
return typeof raw === 'string' ? raw.trim() : null;
};
const matchesTeamMentionText = (text: string): boolean => {
const t = text.toLowerCase();
if (!t.includes(teamLower)) return false;
// Strongest signal: spawn/system prompt format includes: on team "X" (X)
// Use substring checks to avoid regex word-boundary issues with kebab-case names.
if (t.includes(`on team "${teamLower}"`)) return true;
if (t.includes(`on team '${teamLower}'`)) return true;
if (t.includes(`on team ${teamLower}`)) return true;
if (t.includes(`(${teamLower})`)) return true;
return false;
};
const extractTeamFromProcess = (entry: Record<string, unknown>): string | null => {
const init = entry.init as Record<string, unknown> | undefined;
const process = (entry.process ?? init?.process) as Record<string, unknown> | undefined;
const team = process?.team as Record<string, unknown> | undefined;
const raw =
typeof team?.teamName === 'string'
? team.teamName
: typeof team?.team_name === 'string'
? team.team_name
: typeof team?.name === 'string'
? team.name
: null;
return typeof raw === 'string' ? raw.trim() : null;
};
try {
const stream = createReadStream(filePath, { encoding: 'utf8' });
const rl = readline.createInterface({ input: stream, crlfDelay: Infinity });
let teamSeen = assumeTeam;
let taskSeenWithoutTeam = false;
for await (const line of rl) {
const trimmed = line.trim();
if (!trimmed) continue;
try {
const entry = JSON.parse(trimmed) as Record<string, unknown>;
// Team detection (for TaskUpdate without team_name): accept only if we can
// confidently attribute the file to this team.
if (!teamSeen) {
const procTeam = extractTeamFromProcess(entry);
if (procTeam?.toLowerCase() === teamLower) {
teamSeen = true;
}
}
if (!teamSeen) {
const msg = entry.message as Record<string, unknown> | undefined;
const rawContent = msg?.content ?? entry.content;
if (typeof rawContent === 'string' && matchesTeamMentionText(rawContent)) {
teamSeen = true;
}
}
const content = this.extractEntryContent(entry);
if (!Array.isArray(content)) continue;
if (!teamSeen) {
// Check message text blocks for team mention (common in Solo spawn prompts)
for (const block of content) {
if (!block || typeof block !== 'object') continue;
const b = block as Record<string, unknown>;
if (
b.type === 'text' &&
typeof b.text === 'string' &&
matchesTeamMentionText(b.text)
) {
teamSeen = true;
break;
}
}
}
if (teamSeen && taskSeenWithoutTeam) {
rl.close();
stream.destroy();
return true;
}
for (const block of content) {
if (!block || typeof block !== 'object') continue;
const b = block as Record<string, unknown>;
if (b.type !== 'tool_use') continue;
const input = b.input as Record<string, unknown> | undefined;
if (!input) continue;
// Deterministic structured match: any tool whose input references this task+team.
const inputTeam = extractTeamFromInput(input);
const rawTaskId = input.taskId ?? input.task_id;
const inputTaskId = extractTaskIdFromUnknown(rawTaskId);
if (inputTaskId && inputTaskId === taskIdStr) {
// If team is present in the input, require exact match.
if (inputTeam) {
if (inputTeam.toLowerCase() === teamLower) {
rl.close();
stream.destroy();
return true;
}
} else {
// Some agents use TaskUpdate without team_name (common in Solo).
// Only accept when we have a separate team marker for this file.
if (teamSeen) {
rl.close();
stream.destroy();
return true;
}
taskSeenWithoutTeam = true;
}
}
}
if (teamSeen && taskSeenWithoutTeam) {
rl.close();
stream.destroy();
return true;
}
} catch {
// ignore parse errors
}
}
rl.close();
stream.destroy();
} catch {
// ignore
}
return false;
}
private extractEntryContent(entry: Record<string, unknown>): unknown[] | null {
const message = entry.message as Record<string, unknown> | undefined;
if (message && Array.isArray(message.content)) return message.content as unknown[];
if (Array.isArray(entry.content)) return entry.content as unknown[];
return null;
}
private async listSessionDirs(projectDir: string): Promise<string[]> {
try {
const dirEntries = await fs.readdir(projectDir, { withFileTypes: true });
return dirEntries.filter((e) => e.isDirectory()).map((e) => e.name);
} catch {
logger.debug(`Cannot read project dir: ${projectDir}`);
return [];
}
}
private async parseSubagentSummary(
filePath: string,
projectId: string,
sessionId: string,
fileName: string,
targetMember: string,
knownMembers: Set<string>,
precomputedAttribution?: SubagentAttribution
): Promise<MemberSubagentLogSummary | null> {
const subagentId = fileName.replace(/^agent-/, '').replace(/\.jsonl$/, '');
// ── Phase 1: Attribution (first N lines) ──
// Reuse pre-computed attribution when available to avoid re-reading the file.
const attribution =
precomputedAttribution ?? (await this.attributeSubagent(filePath, knownMembers));
if (!attribution) return null;
const targetLower = targetMember.toLowerCase();
if (attribution.detectedMember.toLowerCase() !== targetLower) {
return null;
}
// ── Phase 2: Metadata (stream entire file) ──
// Now that we know the file belongs to this member, collect
// accurate timestamps and message count from the full file.
const metadata = await this.streamFileMetadata(filePath);
const firstTimestamp =
metadata.firstTimestamp ?? attribution.firstTimestamp ?? (await this.getFileMtime(filePath));
const lastTimestamp = metadata.lastTimestamp ?? firstTimestamp;
const startTime = new Date(firstTimestamp);
const endTime = new Date(lastTimestamp);
const durationMs = endTime.getTime() - startTime.getTime();
// Check if the file might still be active (modified recently)
let isOngoing = false;
try {
const stat = await fs.stat(filePath);
const ageMs = Date.now() - stat.mtimeMs;
isOngoing = ageMs < 60_000; // Active within last minute
} catch {
// ignore
}
return {
kind: 'subagent',
subagentId,
sessionId,
projectId,
description: attribution.description || `Subagent ${subagentId}`,
memberName: targetMember,
startTime: firstTimestamp,
durationMs: Math.max(0, durationMs),
messageCount: metadata.messageCount,
isOngoing,
filePath,
};
}
/**
* Phase 1: Scan first ATTRIBUTION_SCAN_LINES lines for member detection signals
* and extract a human-readable description from the first user message.
* Returns null if the file is a warmup session or empty.
*
* Collects ALL detection signals, then selects the best one by precedence
* (process_team > routing_sender > teammate_id > text_mention).
*/
private async attributeSubagent(
filePath: string,
knownMembers: Set<string>
): Promise<SubagentAttribution | null> {
const lines: string[] = [];
try {
const stream = createReadStream(filePath, { encoding: 'utf8' });
const rl = readline.createInterface({ input: stream, crlfDelay: Infinity });
let count = 0;
for await (const line of rl) {
if (count >= ATTRIBUTION_SCAN_LINES) break;
const trimmed = line.trim();
if (trimmed) {
lines.push(trimmed);
count++;
}
}
rl.close();
stream.destroy();
} catch {
return null;
}
if (lines.length === 0) return null;
let description = '';
const signals: DetectionSignal[] = [];
let firstTimestamp: string | null = null;
for (const line of lines) {
if (!firstTimestamp) {
firstTimestamp = this.extractTimestampFromLine(line);
}
try {
const msg = JSON.parse(line) as Record<string, unknown>;
const role = this.extractRole(msg);
const textContent = this.extractTextContent(msg);
// Skip warmup messages
if (role === 'user' && textContent?.trim() === 'Warmup') {
return null;
}
// Extract description from first user message + collect teammate_id signal
if (role === 'user' && textContent) {
if (textContent.trimStart().startsWith('<teammate-message')) {
const parsed = parseAllTeammateMessages(textContent);
if (!description) {
description =
parsed[0]?.summary || parsed[0]?.content?.slice(0, 200) || 'Teammate spawn';
}
// teammate_id identifies the MESSAGE SENDER (e.g. "team-lead"), not the agent
// owning this file. Collected as a signal — higher-precedence sources override.
if (parsed[0]?.teammateId) {
const tmId = parsed[0].teammateId.trim().toLowerCase();
if (tmId.length > 0 && knownMembers.has(tmId)) {
signals.push({ member: parsed[0].teammateId.trim(), source: 'teammate_id' });
}
}
} else if (!description) {
description = textContent.slice(0, 200);
}
}
// Collect text_mention signal (lowest reliability — exact one member name in text)
const textMention = this.detectMemberFromMessage(msg, knownMembers);
if (textMention) {
signals.push({ member: textMention.name, source: 'text_mention' });
}
// Collect routing_sender signal (high reliability — identifies the actual agent)
if (msg.toolUseResult && typeof msg.toolUseResult === 'object') {
const routing = (msg.toolUseResult as Record<string, unknown>).routing as
| Record<string, unknown>
| undefined;
if (routing && typeof routing.sender === 'string') {
const sender = routing.sender.toLowerCase();
if (knownMembers.has(sender)) {
signals.push({ member: routing.sender, source: 'routing_sender' });
}
}
}
// Collect process_team signal (highest reliability — from system init message)
const init = msg.init as Record<string, unknown> | undefined;
const process = (msg.process ?? init?.process) as Record<string, unknown> | undefined;
const team = process?.team as Record<string, unknown> | undefined;
if (team && typeof team.memberName === 'string') {
const memberNameLower = team.memberName.trim().toLowerCase();
if (memberNameLower.length > 0 && knownMembers.has(memberNameLower)) {
signals.push({ member: team.memberName.trim(), source: 'process_team' });
}
}
} catch {
// Skip malformed lines
}
// Early exit: reliable signal found and description extracted — no need to scan further.
// Only process_team and routing_sender trigger this; teammate_id is unreliable (identifies
// the message sender, not the agent) so we keep scanning for better signals.
if (
description &&
signals.some((s) => s.source === 'process_team' || s.source === 'routing_sender')
) {
break;
}
}
if (signals.length === 0) return null;
const best = TeamMemberLogsFinder.selectBestSignal(signals);
if (!best) return null;
return { detectedMember: best.member, description, firstTimestamp };
}
/**
* Select the best detection signal by precedence.
* Signals are collected in file order, so find() returns the earliest occurrence
* of the highest-precedence source.
*/
private static selectBestSignal(signals: DetectionSignal[]): DetectionSignal | null {
for (const source of SIGNAL_PRECEDENCE) {
const match = signals.find((s) => s.source === source);
if (match) return match;
}
return null;
}
/**
* Last-resort member detection from message text.
* Only called when all structured signals (teammate_id, process.team, routing) failed.
* Returns priority 1 (lowest) — only if exactly one known member name appears.
*/
private detectMemberFromMessage(
msg: Record<string, unknown>,
knownMembers: Set<string>
): { name: string; priority: number } | null {
if (this.extractRole(msg) !== 'user') return null;
const text = this.extractTextContent(msg);
if (!text) return null;
// Only attribute if exactly one known member name appears (word-boundary match).
// Avoids false positives when multiple members are mentioned.
const matches: string[] = [];
for (const name of knownMembers) {
const regex = new RegExp(`\\b${escapeRegex(name)}\\b`, 'i');
if (regex.test(text)) {
matches.push(name);
}
}
if (matches.length === 1) {
return { name: findOriginalCase(text, matches[0]), priority: 1 };
}
return null;
}
private extractTextContent(msg: Record<string, unknown>): string | null {
if (typeof msg.content === 'string') {
return msg.content;
}
if (Array.isArray(msg.content)) {
const textParts = (msg.content as Record<string, unknown>[])
.filter((part) => part.type === 'text' && typeof part.text === 'string')
.map((part) => part.text as string);
if (textParts.length > 0) return textParts.join(' ');
}
// Also check message wrapper
if (msg.message && typeof msg.message === 'object') {
return this.extractTextContent(msg.message as Record<string, unknown>);
}
return null;
}
private extractRole(msg: Record<string, unknown>): string | null {
if (typeof msg.role === 'string') {
return msg.role;
}
if (msg.message && typeof msg.message === 'object') {
const inner = msg.message as Record<string, unknown>;
if (typeof inner.role === 'string') {
return inner.role;
}
}
return null;
}
private async parseLeadSessionSummary(
jsonlPath: string,
projectId: string,
sessionId: string,
memberName: string
): Promise<MemberLogSummary | null> {
try {
await fs.access(jsonlPath);
} catch {
return null;
}
const metadata = await this.streamFileMetadata(jsonlPath);
const firstTimestamp = metadata.firstTimestamp ?? (await this.getFileMtime(jsonlPath));
const lastTimestamp = metadata.lastTimestamp ?? firstTimestamp;
const startTime = new Date(firstTimestamp);
const endTime = new Date(lastTimestamp);
const durationMs = endTime.getTime() - startTime.getTime();
let isOngoing = false;
try {
const stat = await fs.stat(jsonlPath);
const ageMs = Date.now() - stat.mtimeMs;
isOngoing = ageMs < 60_000;
} catch {
// ignore
}
return {
kind: 'lead_session',
sessionId,
projectId,
description: 'Lead session',
memberName,
startTime: firstTimestamp,
durationMs: Math.max(0, durationMs),
messageCount: metadata.messageCount,
isOngoing,
filePath: jsonlPath,
};
}
/**
* Stream entire JSONL file collecting only timestamps and message count.
* Lightweight — uses regex to extract timestamp without full JSON parse.
*/
private async streamFileMetadata(filePath: string): Promise<StreamedMetadata> {
let firstTimestamp: string | null = null;
let lastTimestamp: string | null = null;
let messageCount = 0;
try {
const stream = createReadStream(filePath, { encoding: 'utf8' });
const rl = readline.createInterface({ input: stream, crlfDelay: Infinity });
for await (const line of rl) {
const trimmed = line.trim();
if (!trimmed) continue;
messageCount++;
// Fast timestamp extraction without full JSON parse.
// ISO prefix anchor avoids false positives from "timestamp" inside string values.
const ts = this.extractTimestampFromLine(trimmed);
if (ts) {
if (!firstTimestamp) firstTimestamp = ts;
lastTimestamp = ts;
}
}
rl.close();
stream.destroy();
} catch {
// ignore — return whatever we collected so far
}
return { firstTimestamp, lastTimestamp, messageCount };
}
private extractTimestampFromLine(line: string): string | null {
const tsMatch = /"timestamp"\s*:\s*"(\d{4}-\d{2}-\d{2}T[^"]+)"/.exec(line);
return tsMatch?.[1] ?? null;
}
private async probeFirstTimestamp(
filePath: string,
maxLines = ATTRIBUTION_SCAN_LINES
): Promise<string | null> {
try {
const stream = createReadStream(filePath, { encoding: 'utf8' });
const rl = readline.createInterface({ input: stream, crlfDelay: Infinity });
let seen = 0;
for await (const line of rl) {
const trimmed = line.trim();
if (!trimmed) continue;
const ts = this.extractTimestampFromLine(trimmed);
if (ts) {
rl.close();
stream.destroy();
return ts;
}
seen++;
if (seen >= maxLines) break;
}
rl.close();
stream.destroy();
} catch {
// ignore
}
return null;
}
private async getSortTime(filePath: string, timestamp: string | null): Promise<number> {
const resolvedTimestamp = timestamp ?? (await this.getFileMtime(filePath));
const sortTime = Date.parse(resolvedTimestamp);
return Number.isFinite(sortTime) ? sortTime : 0;
}
private async getFileMtime(filePath: string): Promise<string> {
try {
const stat = await fs.stat(filePath);
return stat.mtime.toISOString();
} catch {
return new Date().toISOString();
}
}
}
function findOriginalCase(text: string, lowerName: string): string {
const regex = new RegExp(`\\b(${escapeRegex(lowerName)})\\b`, 'i');
const match = regex.exec(text);
return match ? match[1] : lowerName;
}
function escapeRegex(str: string): string {
return str.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
}