session-scan

What problem does it solve?

This Skill computes deterministic exploratory metrics from Claude session transcripts so contributors can triage, rank, and prioritize sessions for deeper review rather than relying on anecdotal signals or manual inspection.

Core Features & Use Cases

• Provider-scoped discovery and filtering: discover recent sessions via ccrider when available and restrict analysis to a single provider or project to avoid mixed-provider bias.
• Deterministic scoring via canonical script: uses references/compute-metrics.py to compute friction, fingerprint, plugin opportunity, tool profile, file hotspots, and session duration.
• Scalable, safe orchestration: deduplicates against an append-only ledger, spawns one lightweight subagent per session to avoid fetching large transcripts in the main context, supports listing, rescan, and writing scan metadata for trends.
• Use Case: run a scan to surface recent claude-code sessions with high friction scores and identify candidates for manual deep dive and verification.

Quick Start

Run /session-scan --since 7 days ago --provider claude-code to generate a triage table of recent sessions and append results to .claude/session-metrics/metrics.jsonl.