What problem does it solve? Teams accumulate raw learning entries in an inbox file, but without a structured triage process these insights are never classified, evaluated, or converted into reusable knowledge, and risky irreversible changes can happen without approval. ## Core Features & Use Cases - Structured Evaluation Pipeline: Reads and normalizes inbox.md entries, classifies them into problem classes, scores them on an 8-axis rubric, and generates an evaluation report. - Disposition Judgment with Review: Determines promote/defer/reject/duplicate outcomes, checks against existing countermeasures, and runs an adversarial review loop before decisions are finalized. - Autonomous Decisions with HITL Fallback: Auto-confirms items that can be uniquely determined and escalates only genuinely ambiguous items for human approval, then persists results with atomic deferred moves, pruning, and scoped git commits. - Use Case: After several development sessions, run the learning-promote command to triage all accumulated learnings, promote the valuable ones into promoted/ artifacts routed to backlog review, and safely archive the rest. ## Quick Start Ask the agent to run the /agentdev/learning-promote command to evaluate and promote the pending learning entries in the inbox.
