Mentor

A Claude skill for generating structured, official-first, dependency-ordered learning paths for technical topics.

Why this is different

Most learning-roadmap content online is one of three things:

• generic link dumps
• community-heavy resource lists with weak sequencing
• broad study plans that ignore the learner's actual goal

Mentor is designed to be tighter than that:

• official-first when canonical docs exist
• dependency-ordered so the path builds in the right sequence
• practical about what to read now, skim, practice, or bookmark
• adaptive to topic scope, learner background, and time budget
• skill-native for Claude, instead of being a blog post pretending to be a tool

Proof at a glance

• 4 example outputs covering broad and narrow technical topics
• 9 eval cases with machine-checkable assertions
• release-ready mentor.skill asset for direct installation

What It Does

Given a topic like "Terraform", "Google Cloud Run", or "React hooks", Mentor generates:

• A learner profile (inferred or defaulted)
• A dependency-ordered core learning path
• Optional exploration branches
• Self-assessment checkpoints
• An explicit skip-list (avoid for now)
• Follow-on topic suggestions

How It Works

flowchart TD
    A["User asks to learn a topic"] --> B{"Classify topic scope"}
    B -->|Broad| C["4-phase path\n(Mental Model → Core Mechanics → Applied Patterns → Go Deeper)"]
    B -->|Narrow| D["Compressed path\n(Orientation → Core → Applied → Reference)"]
    C --> E{"Infer learner profile"}
    D --> E
    E --> F{"Clarification needed?"}
    F -->|"Would change first 5 resources"| G["Ask 1 short question"]
    F -->|No| H["Build path with defaults"]
    G --> H
    H --> I["Research & verify sources\n(web search for freshness)"]
    I --> J["Apply source ranking\nTier 1 → 2 → 3 → 4"]
    J --> K["Assemble output"]

Source ranking: official docs first, vendor/maintainer material second, official sample repos third, community content only when it fills a real gap.

Output Structure

block-beta
    columns 1
    block:core["Core Path"]
        columns 4
        p1["Phase 1\nMental Model"]
        p2["Phase 2\nCore Mechanics"]
        p3["Phase 3\nApplied Patterns"]
        p4["Phase 4\nGo Deeper"]
    end
    block:branches["Exploration Branches (optional, non-sequential)"]
        columns 3
        b1["Hands-on\nPractice"]
        b2["Architecture\n& Team Usage"]
        b3["Deep\nInternals"]
    end
    block:supporting["Supporting Sections"]
        columns 4
        s1["Checkpoints"]
        s2["Avoid for Now"]
        s3["Next Topics"]
        s4["Navigator's Note"]
    end

Every core resource includes: source tier, engagement mode (Read now / Skim / Hands-on / Bookmark as reference), exact URL, sequencing rationale, and effort estimate.

Repository Structure

flowchart LR
    root["mentor/"] --> skill["SKILL.md\nskill definition"]
    root --> refs["references/"]
    root --> examples["examples/"]
    root --> evals["evals/"]
    refs --> schema["schema.json\nJSON output contract"]
    examples --> ex1["example-output-rust.md"]
    examples --> ex2["example-output-cloud-run.md"]
    examples --> ex3["example-output-terraform-modules.md"]
    examples --> ex4["example-output-react-server-components.md"]
    evals --> evj["evals.json\n9 test cases with assertions"]

• SKILL.md is the source of truth. All behavior rules, output format, source ranking, and anti-patterns live here.
• references/schema.json defines the machine-readable output contract for JSON mode.
• examples/ contains gold-standard outputs demonstrating broad topics, narrow topics, clarification triggers, and user-background adaptation.
• evals/evals.json contains 9 test cases with machine-gradable assertions covering structure, source quality, dedup, mode variety, and more.

Design Principle

Build the shortest credible path to competence from trustworthy sources, while preserving room for exploration.

In practice, that means the skill should feel like a strong staff engineer or teacher handing you the right sequence, not a search engine dropping 20 tabs in your lap.

Target Runtime

This skill is designed for Claude.ai and Claude Code. It uses the Claude skill format (SKILL.md with YAML frontmatter) and relies on Claude's web search capability to verify resource URLs and freshness.

Installation

Option 1: Download the .skill file (recommended)

Download mentor.skill from the latest release, then:

| Platform | Command / Action | |---|---| | Claude Code | claude skill add mentor.skill | | Claude.ai | Open a Project → Settings → Skills → Upload mentor.skill |

Option 2: Clone the repository

git clone https://github.com/ayhammouda/mentor.git ~/.claude/skills/mentor

Option 3: Manual copy

Copy the mentor/ directory into your Claude skills location (e.g., ~/.claude/skills/mentor/).

For more details on installing and managing skills, see the official documentation:

• Claude Code Skills
• Claude.ai Agent Skills

Usage

Ask Claude to learn something:

• "I want to learn Kubernetes"
• "learning path for Terraform modules"
• "teach me Rust, I'm a Go developer"

Mentor activates automatically and generates a structured learning path.

For JSON output, ask explicitly: "Give me a learning path for Docker in JSON format"

Contributing

Contributions welcome! See CONTRIBUTING.md for guidelines on proposing changes, adding examples, and strengthening eval coverage.

License

MIT