construct-card-evaluate

What problem does it solve? Deciding which knowledge cards should be promoted from seed to growing to mature is a judgment-heavy task that drifts when re-implemented in prompts. This Skill delegates that judgment to a single tested Python promotion gate, so promotion decisions stay consistent and reviewable. ## Core Features & Use Cases - Promotion Gate Invocation: Runs construct card evaluate (or the MCP equivalent) against a workspace to produce one PromotionDecision per non-mature card. - Decision Presentation: Groups decisions into promote, hold, and escalate, surfacing the method field so failure-driven escalations (rule-based) are distinguishable from genuine borderline calls (llm-judgment). - Read-Only Evaluation: Proposes decisions without writing lifecycle changes; approved promotions flow through the reviewed curation path. - Use Case: During a curation cycle, ask the agent to check promotions; it runs the gate and returns a summary such as "3 promote, 5 hold, 2 escalate (1 borderline, 1 evaluation failure)" for your review. ## Quick Start Ask the agent to evaluate cards and check which ones are ready for promotion in the current workspace.