MAGIC Agent Skills is now open source! Star on GitHub
MAGIC Agent SkillsMAGIC Agent Skills
Concepts

Concepts

Core concepts that underpin the Linguistic Agent Skills suite.

Pipeline Architecture

The 5-phase model (Scope → Acquire → Analyze → Evaluate → Release), how specialist skills map to each phase, workspace state structure, and the optional Phase 4 Mindset stubs.

Typological Profiling

How URIEL typological distance vectors are used to select transfer sources and predict ML model behavior. Covers key outlier features (polysynthesis, tone, agglutination, root-and-pattern, evidentiality) and the databases used (WALS, Grambank, URIEL, Glottolog).

Joshi Classification

The 6-level resource classification system (Classes 0–5) from Joshi et al. (ACL 2020), with language examples, strategy implications, and the multi-dimensional nature of resource class assessment.

Shared Utilities

The _linguistic_shared/ library — interaction_utils.py and findings_presenter.py — that provide consistent workspace state management and findings presentation across all 18 skills.

Quality Gating

The skill-judge 8-dimension 120-point rubric, per-tier score requirements (entry-point A−, specialist A−, Mindset stub B+), scores for all 18 skills, and the eval methodology.

Was this page helpful?
Edit on GitHub

Last updated on

On this page