Epic: Claude-native telemetry & analysis

Goal

Add a Claude-native telemetry ingestion path so the retrospective can objectively measure how Claude is used (tokens, tools, subagents, skills, MCP, waste, delivery yield), feeding the existing 6 scoring dimensions. Every metric traces to a specific session record — "no AI slop."

Today src/analyzers/tools.ts reads only .logs/tools/*.jsonl from this plugin's own hooks and ignores the authoritative stream under ~/.claude/projects/.

Specification

• Data inventory (primary-source schema): docs/research/claude-data-inventory.md
• Consolidated spec + normalized data contract + roadmap: docs/specs/claude-native-telemetry.md
• Feature PRDs: docs/prd/features/01-08

Capabilities

• #36 — Session ingestion (foundation; normalized data contract)
• #37 — Token & cost accounting
• #38 — Source attribution (tool / subagent / skill / MCP)
• #39 — Waste & efficiency detectors
• #40 — Delivery correlation (shipped-vs-reverted yield, verify loops)
• #41 — Configuration & asset utilization (ghost assets)
• #42 — Durable telemetry ledger (trends past log rotation)
• #43 — Transcript evidence & compact export

Build roadmap (one PR per tier)

1. PR-1 (Tier 1): #36 — parser, normalized model, dedup, subagent/tool-result resolution. Read-only; no scoring change.
2. PR-2 (Tier 2): #37, #38 — accounting + attribution + offline pricing + sprint-window aggregation.
3. PR-3 (Tier 3): #39, #40, #41 — detectors and the delivery/asset signals feeding the dimensions.
4. PR-4 (Tier 4): #42, #43 — durable snapshot store, optional hook ingestion, compact transcript export.

Dimension coverage

Strong: Collaboration Efficiency, Delivery Predictability, Test Loop Completeness, Quality & Maintainability. Decision Hygiene and Security Posture are NOT in ~/.claude/ data — they join the existing decision-log and security analyzers (do not synthesize from token data).

Related

Trend tracking (#18) is served by #42; multi-repo / same-basename project resolution (#19) affects #36.

Start with src/analyzers/tools.ts, then read docs/research/claude-data-inventory.md and docs/specs/claude-native-telemetry.md for the authoritative data and normalized contract. Use #36 as the foundation and review the linked feature PRDs and tier roadmap. Done requires the planned telemetry capabilities to be implemented across the stated tiers without synthesizing Decision Hygiene or Security Posture from Claude data.