What problem does it solve?
Manual harvesting of patterns, conflicts, and learnings from development sessions is time-consuming, inconsistent, and often misses critical insights that could improve your Forge Harness (FH) ecosystem. Long sessions also risk semantic drift, where agent terminology gradually diverges in meaning without detection, leading to inconsistent outputs across your project portfolio.
Core Features & Use Cases
- 8-Step Automated Evolution Pipeline: Runs automatically at session end (or on demand) to extract session patterns, detect collisions with existing skills, validate new skill candidates via adversarial and innovative review, and manage skill lifecycle.
- Semantic Drift Detection: Automatically identifies and blocks terminology divergence across long sessions to maintain consistency in AI-generated outputs.
- Lifecycle Management: Automatically flags stale, unused, or overlapping skills, and maintains memory hygiene to keep your FH ecosystem clean and up-to-date.
- Use Case: After a 3-hour feature development session where your team discovered a new pattern for API error handling, harvest-loop automatically extracts the pattern, checks for conflicts with existing skills, validates a new skill proposal, and updates the shared FH ecosystem so all team projects benefit from the improvement without manual work.
Quick Start
Invoke the harvest-loop skill at the end of your FH field session to automatically extract learnings, validate new skill candidates, and evolve your shared harness ecosystem.
