---
name: explore-new-domain
description: Systematically enter, research, learn, and practice an unfamiliar field by building a concept-and-dependency map, identifying frameworks and strategies, grading evidence, extracting best practices and evaluation criteria, analyzing human and AI failure modes, choosing tools, making novice-safe professional decisions, and running mastery and recursive-validation loops. Use when the user wants to understand or start working in a new domain, requests a knowledge map or learning roadmap, needs current and verified resources, wants sustained tutoring or mastery tracking, lacks expertise to choose among professional options, or wants to turn a vague interest into an evidence-based research and practice plan.
---

# Explore New Domain

Expand the user's cognitive boundary while building independent judgment. Produce a usable map, decision system, and practice loop rather than a fact dump or oversized reading list.

## Route the Request

Choose only the modes needed:

- **Orientation:** Build a compact domain map and immediate entry path. Use by default.
- **Evidence:** Research current knowledge, sources, consensus, disputes, and gaps. Read [references/evidence-protocol.md](references/evidence-protocol.md) before gathering sources.
- **Decision:** Compare professional options with explicit criteria and recommend a default.
- **Mastery:** Tutor, assess, practice, review, or track progress over time. Read [references/mastery-loop.md](references/mastery-loop.md) before teaching.

Combine all four modes for an end-to-end request. Do not force research, persistent files, visual artifacts, or a long tutoring dialogue when the user only needs orientation.

## 1. Establish the Mission

Infer what is safe to infer, then establish:

- Target domain and desired outcome
- Current knowledge and transferable experience
- Time horizon, available effort, and constraints
- Required depth: literacy, competent practice, research, or professional judgment
- A concrete artifact, performance, or decision that can demonstrate progress

Ask only when a missing answer would materially change the path. Otherwise, state assumptions and proceed.

## 2. Build the Domain Map

Map the field before teaching details:

1. Define its boundaries and adjacent fields.
2. Identify foundational concepts and their relationships.
3. Order concepts by genuine prerequisite dependency, not arbitrary sequence.
4. List major frameworks, schools of thought, and strategies.
5. Separate stable fundamentals from contested or fast-changing claims.
6. Mark milestones where the user can demonstrate a real capability.

Prioritize concepts with high explanatory power. Explain specialist terms in plain language and connect each term to a decision or action. Use a visual map only when relationships are difficult to understand in prose.

## 3. Establish the Evidence Base

When the request needs research, current information, recommended resources, or high-stakes accuracy, apply the evidence protocol.

At minimum:

- Define scope, timeframe, and inclusion boundaries.
- Prefer primary and authoritative sources.
- Verify that recommended resources exist, are relevant, and are sufficiently current.
- Distinguish consensus, contextual heuristics, contested views, community signals, and inference.
- Grade evidence quality instead of treating all sources equally.
- Identify contradictions, missing evidence, and questions the current evidence cannot answer.
- Stop gathering when information is sufficient for the current decision or practice task.

Treat popularity and recent community discussion as signals, not proof.

## 4. Extract the Meta-Layer

For the user's actual goal, identify:

- **Frameworks:** reusable structures for thinking
- **Strategies:** approaches for reaching the goal
- **Best practices:** methods supported by evidence or repeated experience
- **Evaluation criteria:** observable standards for judging quality
- **Failure modes:** common ways the human and the AI can fail
- **Tools:** useful software, methods, datasets, communities, and references
- **Tradeoffs:** what each option optimizes and sacrifices

Do not present conventions as universal laws. Tie every recommendation to context, evidence, or a clearly labeled inference.

## 5. Make Novice-Safe Decisions

Do not push an uninformed professional choice back to the novice.

1. Define the decision, outcome, and constraints.
2. Generate 2–4 viable options when alternatives genuinely exist.
3. Select 3–7 criteria relevant to the goal.
4. Weight criteria only when it clarifies priorities.
5. Compare options using the same evidence and assumptions.
6. Recommend a default and explain why it fits.
7. State confidence, important risks, and conditions that would change the recommendation.

For consequential or uncertain choices, seek independent expert perspectives or cross-model consultation when tools and authorization allow. Ask each perspective to expose assumptions, evidence, objections, and confidence. Synthesize with the rubric; do not decide by majority vote and do not invent consultations.

## 6. Design the Learning Path

Divide the path into:

1. **Prerequisites:** knowledge that truly blocks later progress
2. **Core capabilities:** skills directly required by the goal
3. **Optional extensions:** useful but nonessential breadth or specialization

Provide a shortest viable path and an extended path when helpful. Skip material already demonstrated. For every learning unit, include:

- Why it matters
- Prerequisites
- One or two verified core resources when research is available
- A concrete practice task
- A self-check question
- Observable evidence of mastery

Prefer authentic projects and feedback over passive consumption. Make stages produce visible artifacts such as an analysis, experiment, decision memo, explanation, or prototype.

## 7. Run the Mastery Loop When Requested

Use the mastery protocol for sustained learning:

```text
diagnose → decompose → question/model → practice → evaluate
→ diagnose misconception → revise → retrieve later → transfer
```

Do not equate mastery with immediate quiz accuracy. Require a suitable combination of explanation, boundary recognition, authentic performance, error correction, transfer, and delayed recall.

Treat direct explanation and Socratic questioning as tools, not dogma. Let the learner attempt first; escalate from questions to hints to worked examples when needed.

## 8. Validate Recursively

Treat the first answer, plan, or artifact as a draft:

```text
propose → verify facts → surface assumptions → find counterexamples
→ compare alternatives → evaluate by criteria → revise → explain
```

Check separately:

- **Facts:** Are claims supported and current?
- **Logic:** Do conclusions follow from assumptions and evidence?
- **Scope:** Does the result fit the user's context?
- **Adversarial strength:** What would a strong critic challenge?
- **Calibration:** Does confidence match evidence quality?
- **Transfer:** Does the method still work in a different example?

## 9. Control Failure Modes

Cover both sides and attach a detection or prevention mechanism to each material risk.

Common human failures include unclear goals, passive learning, premature specialization, authority bias, overconfidence, resource hoarding, and abandoning feedback loops.

Common AI failures include fabricated facts, false consensus, stale knowledge, hidden assumptions, generic plans, shallow comparisons, and confidently optimizing the wrong objective.

Prefer tests, checkpoints, source verification, counterexamples, reversible experiments, and explicit rubrics over vague warnings.

## Produce the Output

Scale the output to the request. For a full engagement, include:

1. Goal, assumptions, and success evidence
2. Compact domain and dependency map
3. Frameworks, strategies, and current evidence
4. Best practices, evaluation rubric, and failure safeguards
5. Recommended tools and resources with reasons
6. Shortest viable learning-and-practice path
7. One immediate authentic task
8. Self-check and mastery evidence
9. Open questions and the next update point

End with the smallest useful next action. On a later session, begin by testing recall or reviewing the previous artifact, then update the map, evidence, misconceptions, and next step. Persist a learner profile or progress files only when the user asks or the task explicitly requires durable tracking.