ADLC Component Telemetry

Real-time usage tracking across 37 agents, 141 commands, 125 skills, 25 MCPs

Stats·Capabilities·Pipeline·Usage Data·Limits

Usage Summary

Total Invocations

Sum across all sessions

Unique Components

Distinct component names

Component Types

Sessions Tracked

Distinct session dates

Enterprise Capability Map

What the HITL manages — 6 component types powering the AI agent team

👤37

Agents

AI team members mapped 1:1 to market roles

5 opus · 28 sonnet · 4 haiku

e.g. product-owner, cloud-architect, python-engineer

⚡141

Commands

Slash commands automating enterprise workflows

15 categories: finops, itsm, ceremony, aws...

e.g. /adlc, /ceremony:standup, /finops:aws-monthly

🧠125

Skills

Domain knowledge packages loaded on-demand

governance, quality, observability, engineering...

e.g. agent-engineering-7skills, golden-prompt-pattern

🔌25

MCPs

Model Context Protocol integrations (enterprise profile)

AWS (12) · Azure (6) · Atlassian (2) · DevTools (5)

e.g. cost-explorer, atlassian-tools, context7, github

🔒38

Hooks

Deterministic enforcement gates on tool calls

7 lifecycle events · exit 2 = BLOCK

e.g. enforce-coordination.sh, track-component-usage.sh

📋22

Rules

Behavioral governance guardrails (advisory)

governance, engineering, quality categories

e.g. coding-discipline, anti-patterns-catalog

Telemetry Pipeline — What Fires When

7 hook events track component usage across the full session lifecycle into SQLite

PHASE 1Session Init

SessionStartinstruction

SessionStart

→

load-project-context.sh

→

HITL Summary

Creates evidence directories, detects version drift, surfaces DORA + coordination status

PHASE 2Runtime Tracking

User types /cmdcommand

User types /cmd

→

track-component-usage.sh

→

SQLite

UserPromptExpansion captures which /commands the HITL invokes

Skill file loadsskill

Skill file loads

→

track-component-usage.sh

→

SQLite

InstructionsLoaded captures which of 125 skills actually load per session

Agent delegatedagent

Agent delegated

→

track-component-usage.sh

→

SQLite

PostToolUse[Agent] captures specialist delegations with estimated duration

MCP tool calledmcp

MCP tool called

→

track-component-usage.sh

→

SQLite

PostToolUse[mcp__.*] captures calls to 25 enterprise MCP servers

PHASE 3Session Close

Session endsinstruction

Session ends

→

enforce-code-quota.sh

→

Usage Table

Stop event renders component usage table to HITL + warns on zero code changes

agent

skill

command

instruction

mcp

Usage Data

0 records — sorted by invocation count desc

Date	Type	Component Name	Invocations	Outcome	Context
No data yet — run the export script or wait for the next session to log usage.

Known Limitations

💰

Token costs not trackable

Claude Code does not expose API billing to hooks. No per-session cost data available.

⏱

Duration is estimated

Specialist lock-file timestamp diff, ±1s accuracy. Not wall-clock agent runtime.

🔍

Coverage: agents, skills, commands, MCPs, instructions tracked

Read/Edit/Bash tool calls skipped to reduce noise. Hook coverage limited to PostToolUse events.