Product

Status: current product baseline and direction. This document separates implemented behavior from later work.

Setup supports organization and individual workspaces. Their product boundaries and future evolution are defined in audience-modes.md.

What is SkillNet

SkillNet turns an idea or existing knowledge into grounded, traceable training that can take a different form for different learner profiles and states.

It is not only a course catalog or a static LMS with a chatbot attached. It reads manuals, procedures, protocols or a generated source, builds a course and keeps the learner experience separate from the knowledge and objective that must remain stable.

SkillNet is open source and self-hosted. A deployment can begin as an organization workspace or an individual workspace.

The same source and objective can produce different explanations, activities, media and interfaces. The system uses current context and revisable learner evidence; it does not claim to know a person’s fixed learning style.

Roles

Role	What they do
Admin	Uploads documents, reviews generated content, assigns training, sees team progress
Employee	Learns, practices and asks questions. The experience can respond to their profile, current state and support needs

Current learning surfaces

Surface	Purpose
Course	Modules + exercises + evaluation. Structured learning path, generated from company documents
Course tutor	Tutor attached to a course or enrollment. It retrieves course material for source-specific questions and can answer general questions without course citations
Generated media	Podcasts, infographics, slide decks and narrated slide videos when the required providers are configured

Content generation

Current creation paths:

From documents — upload PDF, DOCX, Markdown or TXT material and generate a grounded course.
From an idea — SkillNet creates a clearly marked model-generated source with provenance and builds from it. This is not equivalent to grounding in uploaded company material.
From external clients — the web UI and /ext/v1 use the same authoring services. Optional A2A and MCP adapters call /ext/v1 when their Compose profiles are enabled.

The static v1 pipeline and dynamic v2 schema path coexist. The delivery decision is made per course. Dynamic schemas pass through proposal, node review and validation before learner delivery. See v1 scope, dynamic courses and AI course design.

Exercises

Multiple types are defined by the content itself. Examples include tests, practical cases and real-world tasks (“do this and tell me if it worked”).

Closed generated exercises are prompted to include an explanation. Dynamic Didact activities preserve server-owned source references. Citation coverage is not yet universal across v1 exercises. Answers are evaluated either deterministically (test, true/false, fill_blank) or by an LLM with a rubric (practical_case, dialogue).

Tracking

Learners complete courses. The system records the evidence it can observe:

Enrollments and course progress
Node completion and mastery
Exercise and activity attempts
Learning events and the rendered experience seen by the learner
Skill levels recorded from course mastery or explicit verification

The admin talent surfaces expose people, assigned courses, progress and recorded skills with their source courses. Completion, mastery and skill remain separate claims. End-to-end lineage from a skill to an attempt, rendered material and source remains active work.

Adaptation

SkillNet separates the stable course contract from the experience served to one learner.

Static delivery: generated Markdown and exercises remain the compatibility path.

Dynamic delivery: a validated course schema can produce per-node episodes using grounded course knowledge, learner profile, current state and an approved component catalog. The controlled runtime can vary the explanation, example, activity, support, medium and interface without changing the objective or evidence requirement. Equivalent learner inputs may share a render; adaptive episodes and multi-agent review are optional features and are disabled by default.

Learner memory: editable memory currently personalizes the tutor. Lesson generation uses declared preferences, learner state and bounded event projections. Using free-form memory to steer shared lesson renders remains future work. This is distinct from immediate intent.

Adaptive regeneration: detecting weak content across many learners and proposing a grounded revision remains future work.

Learning Loop

The implemented loop records evidence from each interaction:

Learner takes course
    |
    v
Experience and attempts recorded
    |
    v
Progress, mastery and skills updated through their own rules
    |
    v
Course-specific tutor and explain flows retrieve course context
    |
    v
Admin sees progress and recorded skills
    |
    v
(Future) evidence supports reviewed changes to experience or content

Recording events is implemented. Proving that an adaptation improves learning and automatically changing the course from aggregate evidence are separate roadmap outcomes.

Living Content

Company documentation changes. SkillNet is designed to treat sources as living rather than static. The following behavior is a product horizon, not the current end-to-end workflow:

When a document is re-uploaded, the system detects what changed
Affected courses are flagged for review
The admin decides whether to regenerate or keep the current version
Employees see a version indicator so they know if their training is current

This lets SkillNet stay in sync with the company’s knowledge instead of generating content only once.