algebench

Interactively build a research-grounded mathematical proof or derivation — research the topic on the web (opt-in per source), confirm the derivation path, generate CAS-verified proof JSON via the algebench-proof-anim pipeline, save it under proofs/domains, and open it on the /prove page. Pass --research to stop at a copy-paste research summary + recommended prompt and (optionally) open a pre-filled Derive tab.

Check an AlgeBench lesson for semantic-graph prebaking — validate which baked graphs are still correct, identify which steps need (re)baking, and bake them into the JSON so the server doesn't derive them at request time. Pass `enrich` to also fill node metadata (descriptions, units, dimensions, emoji, domain) in a second LM pass.

Research a mathematical derivation with the user, generate a shareable AlgeBench proof animation JSON, and launch the app to show it rendering. Interactive and AskUserQuestion-driven at every decision point.

Check deployment status and deploy AlgeBench — Render (staging and prod) and the Hugging Face Space (prod mirror). Production is a choose-host (Render/HF/both) and choose-source (current branch/main/staging) flow.

Release workflow for AlgeBench. Reconciles the VERSION file with the prod git tag, tags a new release on main, publishes a GitHub release, then bumps the VERSION file to the next dev version via a PR. Covers version discovery, tag naming conventions, merge checks, and GitHub release creation.

Launch or restart AlgeBench, open it in Chrome browser tools, and debug UI behavior with console, DOM, screenshots, and interactions.

Update the pinned gemini-live-tools dependency to a PR branch, named branch, version tag, or latest release and verify the installed package.

Triage the GitHub issue backlog and code scanning alerts for the current repo — list, filter, prioritize, and pick issues or alerts. Supports --type filtering by label, --code-scanning for CodeQL/security alerts, --prioritize to mark issues as high priority, and --pick to select and set up a branch for the top item.

Scene Builder for the lesson builder pipeline. Takes a single scene outline and produces complete AlgeBench scene JSON with all steps, elements, sliders, proofs, and markdown in one pass.

Orchestrator for the multi-agent lesson builder pipeline. Coordinates research, pedagogy, design, scene building, validation, and evaluation to produce complete AlgeBench lesson JSON.

Generate or update the AlgeBench lesson JSON Schema by analyzing existing scenes, renderer code, scene-builder skill docs, and proof model documentation. Produces schemas/lesson.schema.json.

Validate an AlgeBench lesson or scene JSON file against the lesson schema. Reports structural errors, missing fields, invalid types, bad expressions, and proof issues. Fixes errors when possible.

Lesson Designer for the lesson builder pipeline. Synthesizes research and pedagogy outputs into concrete scene-by-scene outlines with step details, slider plans, proof placement, and element specifications.

Pedagogical Evaluator for the lesson builder pipeline. Reviews a complete lesson JSON for pedagogical quality, flow, consistency, and engagement. Returns a verdict with actionable feedback.

Pedagogy Expert for the lesson builder pipeline. Designs the pedagogical approach, learning arc, scaffolding strategy, and proof placement for a lesson topic.

Research Agent for the lesson builder pipeline. Gathers mathematical facts, definitions, theorems, proofs, applications, and citations for a given topic. Produces a structured research brief.

Syntax Validator for the lesson builder pipeline. Validates assembled lesson JSON using schema and content validation scripts. Auto-fixes common errors and reports remaining issues.

Audit expression sandbox coverage. Scans all scene JSON files for expressions and verifies each is properly handled by the math.js sandbox or the JS trust verification system. Run this before merging any PR that adds or modifies scene files.