What problem does it solve? Repeated manual procedures, corrections, and re-investigations scattered across past AI coding sessions are hard to spot, so opportunities to turn them into reusable skills, rules, or references get lost. ## Core Features & Use Cases - Cross-agent log mining: Scans both Claude Code (~/.claude/projects) and Codex (~/.codex/sessions) session logs to find work repeated across two or more sessions. - Classification and dedup: Sorts each candidate into skill, operating rule, memory, or reference/script, and checks it against the existing asset catalog to avoid duplicates. - Ranked report-only output: Produces a priority-ranked (high/medium/low) report with evidence and a recommended next flow, without creating or deploying anything. - Use Case: Ask it to review the last 14 days of sessions; it might surface that you repeatedly rebuild the same release checklist by hand and rank it as a high-priority skill candidate to hand off to the skill-authoring flow. ## Quick Start Ask the agent to mine your recent session logs for asset candidates, for example: "セッションログから資産化候補を出して。直近2週間で。"
