Repository navigation
First public release of LEAP Framework β a state-driven agentic tutoring runtime for AI agents, exposed over MCP.
Runtime
- MCP server over stdio exposing 58 tools, with SQLite persistence and an append-only event log
- Server-side State Guard β the only component that may authorise a state transition
- Forgetting-aware BKT mastery estimation, computed server-side and never emitted by the host LLM
- Evidence stages
estimated β practiced β demonstrated β retained β transferred, with regression - FSRS-based review scheduling (rating 1β4)
- Rule-based pedagogical policy with strategy/action separation, interleaving, and learner-requested session overrides
- Misconception tracking, learning-friction signals, transfer probes and learning-outcome metrics
- Six pluggable seams declared as
Protocols and resolved through a plugin registry
Internationalisation
- Message-catalogue layer (
zh-CN,en) for learner-facing text localeconfiguration plusLEAP_LOCALEenvironment override- Tests enforcing identical key sets and placeholders across catalogues
Tooling
security_scan.pyβ secret and PII scanner with a CI gateverify_example.py/verify_example_runtime.pyβ static and headless-runtime compliance checksapply_leap_bridge.pyβ host-bridge injector with--check,--repairand--revertspec_coverage.pyβ spec-versus-code auditor- 260 tests, green on Python 3.11 / 3.12 / 3.13
Examples
- Five single-file teaching pages: programming, mathematics, computer science, physics, logic
- Each verified statically and in headless Chromium, and each demonstrating
LEAP.hydrate()/LEAP.drainOutbox()
Documentation
- Bilingual READMEs plus Japanese, Korean, French, Spanish, Russian and Arabic
- Architecture, integration, examples and localization guides
- Contributing guide, security policy and code of conduct
The authoritative framework design specification and visual design system are maintained by the project author separately and are not part of this repository. The implementation here follows those documents.