diff-context-baseline

What problem does it solve?

It resolves the ambiguity of whether an implementation followed the planned intent by comparing the prepare-time context baseline against post-epic implementation outcomes and converting divergences into structured, tiered learnings.

Core Features & Use Cases

• Baseline vs outcome diffing: Loads context artifacts captured by prepare and compares them against milestone and optional arbiter verdict evidence from the implementation phase.
• Tiered learning classification: Routes each meaningful divergence into Tier 1 (ADR-worthy), Tier 2 (enrichment-worthy), or Tier 3 (addition-worthy) so follow-on skills know what to propose and where it matters most.
• Two-level KB-aligned taxonomy: Emits a learning_category + sub_category taxonomy aligned to the core memory knowledge base structure, replacing the prior single dimension approach, and produces context-diff.yaml for downstream enrichment drafting.
• Knowledge-extractor stage workflow: Designed as the first of three knowledge-extractor skills, feeding results into draft-enrichment-proposals.

Quick Start

Use the diff-context-baseline skill to generate context-diff.yaml by diffing context_baseline_path against milestone_verdicts_paths (and optional arbiter_verdicts_paths) while writing results under output_base using stm_evidence_root for taxonomy-justified routing.