The problem
AI agents can produce plausible-looking lesson materials quickly, but quality varies widely.
Standards alignment, grade-level rigor, curriculum coherence, and classroom usability are hard to guarantee from a single prompt.
What we’re building
Agent Skills
Our Agent Skills package the instructions, references, and guardrails an agent needs to reliably complete specific K-12 teaching workflows.
Each skill is cross-platform, model-agnostic, and produces stronger results when paired with Learning Commons Knowledge Graph.
| Skill | Description | Development partner |
|---|
| K-12 Lesson Planning | Builds classroom-ready, standards-aligned lesson plans, optionally aligned to a teacher’s curriculum. | Anthropic ↗ |
| K-12 Lesson Differentiation | Adapts an existing lesson into tiered versions (below / at / above proficiency level) and for specific student needs, keeping core content consistent across tiers. | Anthropic ↗ |
Rubrics
This is part of a larger evaluator harness that Learning Commons plans to
publish in full at a later date. Manual use instructions are included for now.
| Field | Description |
|---|
id | Unique criterion identifier (e.g., P1, R3) |
bucket | Top-level category: P (Pedagogy), R (Rigor), O (Output/Formatting), or M (Model Scaffolding) |
criterion | Short name for the criterion |
what_pass_requires | Specific and scoreable condition that constitutes a pass |
conditional | If non-empty, the criterion applies only when this condition is met (e.g., K-5, ELA-Gr8+) conditional criteria are applied only when the specified condition is met; otherwise, they are skipped (not failed).
|
notes | Rationale or design notes |
Track per-criterion pass rates across a prompt suite rather than relying on
aggregate scores alone.Since criteria score independently, a failing “Rigor” criterion simply tells
you something specific about cognitive demand – it does NOT mean that the
output is “bad”.
Scope and limitations
Remember that LLM outputs can vary across runs, especially on borderline
cases. We recommend keeping a human in the loop and treating outputs as
directional signals vs. definitive classroom decisions.
| Limitation | Details |
|---|
| Validated workflows | Initial skills focus on K-12 lesson planning and differentiation across math, ELA, science, and social studies. |
| Knowledge Graph dependency | Skills work without Knowledge Graph but produce the strongest outputs when connected to standards and curriculum data. |
| Curriculum coverage | Curriculum-specific branches cover supported materials such as Illustrative Mathematics and OpenSciEd; other curricula fall back to general best practice. |
Agent Skills outputs should not be used for high-stakes applications like
grading, assessment, or placement decisions without human review.
