Forge — Transcript Knowledge Miner

What it does

Forge scans session transcripts to extract structured knowledge: decisions, learnings, failures, and recurring patterns. It can run automatically at session end or be invoked manually to produce candidate learning files that are scored by confidence and queued for human review or auto-promotion.

When to use it

Use Forge after interactive sessions when you want to capture actionable knowledge from conversations: meeting takeaways, postmortems, design decisions, or debugging lessons. Trigger via session-end hooks or run manually with the provided CLI flags to process pending transcripts.

What's included

• Scripts: present (scaffolded CLI hooks and promote scripts)
• References: includes pattern definitions and uncaptured-lesson patterns for matching (references/uncaptured-lesson-patterns.md)
• Instructions: step-by-step guidance for identifying transcripts, extracting candidate learnings, writing forge output files, indexing, and promoting high-confidence items.

Compatible agents

Best used with agent runtimes that support CLI hooks and local file-based skills (ao CLI / agentops style workflows). Works with agents that can run shell hooks and manage .agents/ state (typical for local agent runtimes and Codex-based orchestrators).