Анализ метрик сессий Claude Code

data

session-scan is a Claude Code skill that computes deterministic metrics for Claude Code sessions: it discovers them via ccrider MCP, filters out trivial ones (fewer than 10 messages), and calculates friction, opportunity, and fingerprint scores to triage and prioritize sessions for deeper review. The skill enforces strict subagent delegation — the main context only handles discovery, deduplication, and coordination, while each haiku subagent fetches messages for exactly one session via `get_session_messages` and runs the `compute-metrics.py` scorer. Results are appended to an append-only `.claude/session-metrics/metrics.jsonl` ledger and displayed as a table sorted by friction score descending, with high-scoring sessions flagged as Tier 2 candidates for `/session-deep-dive`. Supported flags include `--since`, `--project`, `--limit`, `--list`, and `--rescan`. Requires ccrider MCP to be installed.