Claude-native telemetry: configuration & asset utilization (ghost assets)

Problem

Installed agents, skills, slash-commands, and MCP servers consume context budget on every prompt whether used or not. "Ghost assets" (loaded but never invoked) are silent recurring waste that token totals never reveal.

Capability

Cross-reference the configured/installed asset inventory against what actually fired during the sprint, and report utilization gaps.

Behaviour

• Inventory assets from ~/.claude/agents/, skills/, commands/, MCP servers in settings.json, project CLAUDE.md includes.
• Build the invocation set from attribution (which agents/skills/commands/MCP methods fired).
• Report ghost assets (installed, never invoked) and MCP loaded-vs-invoked coverage.
• Where estimable, attach recurring per-prompt token cost to prioritize pruning.

Acceptance criteria

• Installed-but-unused skill flagged; installed-and-used skill not flagged.
• MCP loaded-vs-invoked report lists configured-but-unused methods.
• Each ghost-asset finding cites the config source declaring it.

Dimension: Collaboration Efficiency. Tier 3. Depends on #36, #38. PRD: docs/prd/features/06-config-asset-utilization.md.

Read docs/prd/features/06-config-asset-utilization.md first, then inspect the inventories under ~/.claude/agents/, skills/, and commands/, MCP settings in settings.json, and project CLAUDE.md includes. Trace the attribution data for agents, skills, commands, and MCP methods. Done means unused installed assets and configured-but-unused MCP methods are reported with their config sources, while used assets are not flagged.