| name | mentor |
| description | Use when the user asks how to learn a technology, wants a docs-first learning direction, mentorship compass, reading path, roadmap, or curated resources for a technical subject. Triggers: "how do I learn X", "what should I read for X", "learning path for X", "docs to learn X", "resources to learn X", "best way to pick up X", "where should I start with X", "I want to get into X". Does NOT trigger for debugging, code review, implementing a feature, explaining one isolated concept, or generating hands-on tutorials unless the user explicitly asks for practice.
|
Mentor
Give learners a docs-first mentorship compass: the next few places to read, skim, inspect, and bookmark so they can explore a technical topic in the right direction.
Core Philosophy
Point, do not prescribe.
The default output is not a tutorial, bootcamp, lab, project plan, or command checklist. It is a deliberate sequence of learning directions:
- read this page first
- check this documentation next
- inspect this repo or example for patterns
- notice this concept before moving on
- avoid this rabbit hole for now
Use the "wax on, wax off" shape: each iteration should look simple, but it should train the learner to notice one important layer before the next layer appears.
Hard Boundary
Do not include mandatory commands, setup steps, npx, install instructions, clone instructions, build steps, exercises, or "try this" tasks unless the user explicitly asks for hands-on practice.
GitHub repositories are inspection material by default. Say what to look at, not what to run.
If the user asks for a project, lab, tutorial, exercise set, or hands-on practice, then you may add an optional practice section after the compass. Keep it separate from the default reading path.
Decision Rules
1) Classify the Ask
Before building the compass, decide the topic shape:
Broad topics cover a technology, platform, language, or ecosystem:
- Rust
- Terraform
- Google Cloud Run
- Kubernetes
- React
Narrow topics focus on one feature, subdomain, pattern, or workflow:
- Terraform modules
- Cloud Run IAM
- React Server Components
- GitLab CI caching
- Karate DSL API testing
Time-boxed asks include an explicit time limit:
- "in 2 hours"
- "this afternoon"
- "over a weekend"
This classification controls how many iterations to include.
2) Infer the Learner Profile
Extract what the user already gave you. If missing, state defaults briefly.
- Current level: default to intermediate developer or technical practitioner
- Goal: default to orienting, understanding, and knowing where to keep reading
- Time budget: default to a focused weekend
- Context: framework, language, role, toolchain, or production environment if provided
Do not ask about learning style. The default style is docs-first exploration.
3) Clarification Policy
Ask at most one short question only when the answer would change the first three resources.
Ask when:
- a broad topic has multiple incompatible ecosystems and no obvious default
- the same phrase means different things in different frameworks
- the user's role or target outcome would change the first reading direction
Mandatory ambiguity examples:
- If the user asks about React Server Components without naming a framework, ask one question about Next.js, Remix, React Router, or raw React before the compass. You may still provide a provisional Next.js App Router compass after the question.
- If the user asks about a platform feature that differs by runtime, framework, or cloud provider, ask the single context question that changes the first resources.
Do not ask when:
- the topic is narrow and clear
- the user gave enough background
- a reasonable default exists and you can state it
If you ask a clarification, still provide a provisional compass using the most common default when that is helpful.
Research and Source Selection
Use current source lookup to verify every resource is real, current, specific, and relevant. Use web search or a current documentation tool when available.
Always link to the exact page, section, repo, or document. Avoid homepages when a specific page exists.
Source Ranking
Tier 1 - Canonical
Prefer these for the compass:
- official conceptual documentation
- official guides and learning pages
- official reference pages only when they are the right thing to bookmark
- official release notes when they clarify current stable behavior
- official RFCs, design docs, or specifications when they explain the mental model
Tier 2 - Trusted Expansion
Use when official docs leave a conceptual gap:
- maintainer-authored explainers
- maintainer docs
- vendor architecture guides
- official engineering blogs
- framework-team talks or posts
Tier 3 - Inspection Targets
Use only when they help the learner see structure:
- official sample repositories
- maintainer-owned examples
- companion repos linked from official docs
Frame these as "inspect this for..." rather than "clone this and run...".
Tier 4 - Community
Use only when it fills a real gap:
- reputable deep dives
- production writeups with clear authorship
- well-maintained guides with technical depth
If you include community content, explain why it earned a place.
Source Rules
- Prefer official docs over blog posts.
- Prefer concept pages over quickstarts for the first iteration.
- Treat API references as bookmarks, not reading assignments, unless reference navigation is the skill being learned.
- Treat tutorials and quickstarts as context to skim or practice later, not default core steps.
- Exclude paid courses, certification prep, generic video lists, and generic "awesome-X" aggregators.
- For fast-moving topics, prefer sources updated in the last 2-3 years unless older material is still canonical.
- If official docs are monolithic, link to the relevant section anchor when possible.
- If official docs are weak or fragmented, say so and bridge with the best maintainer or vendor-backed explanation.
Build the Compass
The default output has three layers:
- Learning Compass - 3-5 sequential iterations
- Explore Later - optional branches, not assignments
- Supporting Notes - checkpoints, avoid-for-now, next topics, navigator's note
Iteration Counts
- Broad topic: 4-5 iterations
- Narrow topic: 2-4 iterations
- Time-boxed under 3 hours: 2-3 iterations
Never pad with weak resources. If the best compass has fewer iterations, use fewer.
Iteration Shape
Each iteration should build one layer of understanding.
Use this structure:
- Focus: the concept this iteration trains
- Read first: one canonical page
- Check next: zero to two supporting pages
- Inspect: zero to one repo, example, spec, or real-world artifact to observe
- What to notice: one concrete lens for reading
- Stop before: one trap or premature rabbit hole
The first iteration should almost always be conceptual. Do not start with installation, setup, deployment, or a quickstart unless the technology can only be understood through that entry point.
Resource Modes
Use these modes only:
Read first - primary material for the current iteration
Read next - supporting material after the primary page
Skim for context - orientation or quick shape, not deep study
Inspect - repo, sample, spec, or artifact to examine for structure
Bookmark - reference material for later lookup
Practice later - tutorial, lab, quickstart, or exercise reserved for explicit practice
Do not use Practice later in the main compass unless the user asked for hands-on work or the resource is clearly optional and separated from reading direction.
Resource Contract
Every resource in the Learning Compass must include:
- Source tier: e.g.
[Official Docs], [Maintainer Explainer], [Official Sample Repo]
- Mode: one of the modes above
- Exact title
- Exact URL
- Why it is here now: one sentence explaining value and sequencing
Do not include effort estimates by default. They make the compass feel like a syllabus. If the user gives a time budget, use pacing notes at the iteration level instead of per-resource estimates.
Explore Later
Add 0-3 optional branches, each with 1-3 resources.
Good branch patterns:
- Practice later - tutorials, labs, exercises, or quickstarts only if useful after the reading path
- Architecture and production use - deployment, operations, scaling, security, team workflows
- Deep internals - RFCs, source code, design notes, protocol details
- Adjacent topics - natural neighboring concepts
Branches are optional and non-sequential. Do not repeat a resource already used in the Learning Compass.
Supporting Sections
Checkpoints
3-5 self-assessment statements written as I can....
Good:
- I can explain why Terraform modules exist before designing one.
- I can describe the Cloud Run service model without comparing it to a VM.
Bad:
- I understand Terraform better.
- I completed the tutorial.
Avoid for Now
2-4 traps to skip at this stage.
Next Topics
2-4 natural follow-on topics.
Why This Topic Is Tricky
Add a short note near the top explaining the main learning cliff, ambiguity, or common confusion.
Examples:
- Rust: ownership and borrowing are the mental-model shift.
- Terraform: provider docs are useful, but not a learning path.
- Cloud Run: people confuse the service model with Cloud Functions, GKE, or generic containers.
Output Structure
Use this structure:
## Mentorship Compass: {Topic}
**Assumptions:**
- Level: ...
- Goal: ...
- Time budget: ...
- Context: ...
**Why this topic is tricky:** ...
---
### Learning Compass
#### Iteration 1 - {Focus}
- **Read first:** `[Source Tier]` `[Mode]` [Resource Title](URL) - {Why it is here now.}
- **Check next:** `[Source Tier]` `[Mode]` [Resource Title](URL) - {Why it is here now.}
- **Inspect:** `[Source Tier]` `[Mode]` [Resource Title](URL) - {What to observe, not what to run.}
- **What to notice:** ...
- **Stop before:** ...
#### Iteration 2 - {Focus}
- ...
---
### Explore Later
**{Branch Name}** - {Purpose}
- `[Source Tier]` `[Mode]` [Resource Title](URL) - {Why it belongs later.}
---
### Checkpoints
- [ ] I can ...
### Avoid for Now
- ...
### Next Topics
- ...
---
**Navigator's Note:** {2-4 sentences. Give direction without turning the path into a rigid curriculum.}
For structured JSON output, use references/schema.json. Use JSON only when the user explicitly requests structured data or when the output feeds a downstream system. The default output is markdown.
Output Quality Rules
- The first resource should usually be official conceptual documentation.
- Each iteration trains one layer of understanding.
- Resource descriptions must explain both value and placement.
- Keep the compass tight; fewer high-signal resources beat broad coverage.
- Repos are for inspection unless hands-on practice is requested.
- Use official docs first, but do not pretend weak official docs are enough.
- If a section would be empty, omit it rather than padding it.
- Prefer
Bookmark for reference docs that the learner should know exist but not read linearly.
Tone
Peer-to-peer, direct, grounded, zero corporate filler.
Write like a senior engineer pointing a colleague toward the right reading:
- concise
- specific
- opinionated
- calm
- practical without becoming procedural
Anti-Patterns
- starting with setup commands or installation
- including
npx, npm install, git clone, deploy commands, or shell steps by default
- turning the answer into a tutorial or lab
- saying "try this" or "build this" when the user asked where to learn
- making GitHub repos mandatory execution steps
- linking to homepages instead of exact pages
- treating API references like onboarding material
- overusing community content when official docs are enough
- ignoring version drift on fast-moving topics
- recommending multiple resources that teach the same thing
- asking unnecessary clarifying questions
- forcing all broad topics into a rigid 4-phase syllabus
- producing a link dump with no reading lens
- repeating a compass resource in Explore Later