Mines your Claude transcripts for decisions, learnings, failures, and patterns, then stages them for promotion to your knowledge base. Runs automatically via SessionEnd hook or manually with /forge on specific transcripts. The interesting bit is the tier system: raw extractions land in .agents/forge/ at Tier 0, then get promoted to .agents/learnings/ only after human review or hitting confidence thresholds. Think of it as the capture half of a memory system. It tracks 26 known "uncaptured lesson patterns" and reports progress on catching them. If you're building up a knowledge base from sessions rather than relying on ephemeral context, this is the funnel that feeds it.
npx skills add https://github.com/boshu2/agentops --skill forge