Run any Skill in Manus with one click

Get Started

token-data-sources

Stars34

Forks8

UpdatedMarch 13, 2026 at 19:04

When debugging why the analysis page shows missing or zero token data, or when adding new cost/usage visualizations.

Installation

Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.

Run Skill in Manus

Source

justinchuby

justinchuby/flightdeck

View GitHub Repository View Creator Repositories

Download

Run Skill in Manus

Related occupationsSOC

Based on SOC occupation classification

Software DevelopersComputer and Mathematical Occupations·SOC 15-1252

SKILL.md

readonly

name	token-data-sources
description	When debugging why the analysis page shows missing or zero token data, or when adding new cost/usage visualizations.

Analysis Page Token Data Sources

When to use: When debugging why the analysis page shows missing or zero token data, or when adding new cost/usage visualizations.

Two Token Data Paths

The analysis page has two independent data sources for token information. Understanding which is active and when is critical.

Path 1: Live Agent Objects (In-Memory)

Source: AnalysisPage.tsx reads agent.inputTokens / agent.outputTokens from the app store
Populated by: WebSocket agent:usage events → appStore updates agent objects in memory
Works when: Session is active and agents are running
Fails when: Session is inactive (agents gone from memory → all zeros → "Waiting for token data...")
Used for: Time-series token usage chart, total token counts

Path 2: DB-Persisted Cost Records

Source: CostBreakdown.tsx fetches /costs/by-agent and /costs/by-task API endpoints
Populated by: CostTracker.recordUsage() → writes to task_cost_records DB table
Works when: Data has been recorded to the DB (persists across sessions)
Fails when: The dagTaskId gate blocks recording (see below)
Used for: Per-agent and per-task cost attribution breakdown

The dagTaskId Gate Bug

In AgentManager.ts, token usage recording has a gate condition:

// packages/server/src/agents/AgentManager.ts ~line 725
if (dagTaskId && agent.parentId) {
  tracker.recordUsage(agent.id, dagTaskId, ...);
}

This requires BOTH:

dagTaskId — agent must have an active DAG task
agent.parentId — agent must have a parent (not be the lead)

Problem: The lead agent (which has no parentId) never records usage. And child agents only record when they have an active DAG task. This means task_cost_records is often empty → "No token attribution data yet."

Meanwhile, emit('agent:usage') at line ~730 fires unconditionally, so the time-series chart works fine.

Fix Plan

Remove the parentId requirement: The lead's token usage should be tracked too
Fall back to a synthetic task ID: When no dagTaskId exists, use agent.id or a session-level bucket
Add DB fallback for inactive sessions: When live agents aren't available, fetch historical data from /costs/by-agent instead of reading from in-memory agent objects

Key Files

File	Lines	Role
`packages/server/src/agents/AgentManager.ts`	~721-731	Token usage wiring — emit (unconditional) vs record (gated)
`packages/server/src/agents/CostTracker.ts`	102-154	`recordUsage()` — writes to task_cost_records
`packages/server/src/agents/CostTracker.ts`	157-182	`getAgentCosts()` — reads aggregated costs
`packages/server/src/routes/lead.ts`	263-296	`/costs/by-agent` and `/costs/by-task` API endpoints
`packages/web/src/components/AnalysisPage/AnalysisPage.tsx`	84-86	Token data from live agent objects
`packages/web/src/components/TokenEconomics/CostBreakdown.tsx`	34-37	Fetches DB-backed cost data

More from this repository

same repository

adapter-architecture-pattern

justinchuby/flightdeck

How the multi-backend adapter system works — AgentAdapter interface, ACP protocol, session resume, and how to add new provider adapters

2026-03-1534

agent-spawn-provider-inheritance

justinchuby/flightdeck

When debugging why an agent has the wrong provider/model, or when modifying agent creation or delegation flows.

2026-03-1334

group-chat-adoption

justinchuby/flightdeck

How to get agents to use group chats for peer coordination in flightdeck-based crews. Covers hub-and-spoke anti-pattern, auto-grouping triggers, and file-lock-linked groups. Use when setting up multi-agent crews with 3+ agents or diagnosing why agents aren't collaborating directly.

2026-03-1334

activity-feed-server-filtering

justinchuby/flightdeck

When fetching from the `/coordination/activity` API and you only need specific action types (e.g., `progress_update`, `task_completed`, `delegated`).

2026-03-1334

activity-ledger-patterns

justinchuby/flightdeck

When fetching activity data for UI display, or debugging why activity-based features show empty/stale data.

2026-03-1334

architecture-decisions

justinchuby/flightdeck

Architecture decisions and patterns from Flightdeck development (Phases 2–4). Covers feature architecture, state management, component patterns, and API design.

2026-03-1334

Analysis Page Token Data Sources

When to use: When debugging why the analysis page shows missing or zero token data, or when adding new cost/usage visualizations.

Two Token Data Paths

The analysis page has two independent data sources for token information. Understanding which is active and when is critical.

Path 1: Live Agent Objects (In-Memory)

Source: AnalysisPage.tsx reads agent.inputTokens / agent.outputTokens from the app store

Populated by: WebSocket agent:usage events → appStore updates agent objects in memory

Works when: Session is active and agents are running

Fails when: Session is inactive (agents gone from memory → all zeros → "Waiting for token data...")

Used for: Time-series token usage chart, total token counts

Path 2: DB-Persisted Cost Records

Source: CostBreakdown.tsx fetches /costs/by-agent and /costs/by-task API endpoints

Populated by: CostTracker.recordUsage() → writes to task_cost_records DB table

Works when: Data has been recorded to the DB (persists across sessions)

Fails when: The dagTaskId gate blocks recording (see below)

Used for: Per-agent and per-task cost attribution breakdown

The dagTaskId Gate Bug

In AgentManager.ts, token usage recording has a gate condition:

// packages/server/src/agents/AgentManager.ts ~line 725 if (dagTaskId && agent.parentId) { tracker.recordUsage(agent.id, dagTaskId, ...); }

This requires BOTH:

dagTaskId — agent must have an active DAG task

agent.parentId — agent must have a parent (not be the lead)

Meanwhile, emit('agent:usage') at line ~730 fires unconditionally, so the time-series chart works fine.

Fix Plan

Remove the parentId requirement: The lead's token usage should be tracked too

Fall back to a synthetic task ID: When no dagTaskId exists, use agent.id or a session-level bucket

Add DB fallback for inactive sessions: When live agents aren't available, fetch historical data from /costs/by-agent instead of reading from in-memory agent objects

Key Files

File

Lines

Role

packages/server/src/agents/AgentManager.ts

~721-731

Token usage wiring — emit (unconditional) vs record (gated)

packages/server/src/agents/CostTracker.ts

102-154

recordUsage() — writes to task_cost_records

packages/server/src/agents/CostTracker.ts

157-182

getAgentCosts() — reads aggregated costs

packages/server/src/routes/lead.ts

263-296

/costs/by-agent and /costs/by-task API endpoints

packages/web/src/components/AnalysisPage/AnalysisPage.tsx

84-86

Token data from live agent objects

packages/web/src/components/TokenEconomics/CostBreakdown.tsx

34-37

Fetches DB-backed cost data