Research begins with a question — but the question itself must be discovered. Most researchers waste months wandering before finding their direction. This skill replaces wandering with structured crystallization.
Goal-Driven Requirement Refinement Engine for Research
Transforms fuzzy research intent into a crystallized North Star statement and structured ResearchBrief through adaptive dialogue and on-demand investigation. Not a chatbot that asks "what do you want to research?" — a structured coaching system that discovers the answer with you.
- 🎯 Adaptive routing — detects how much context you already have (cold/warm/hot) and adjusts depth accordingly. No wasted questions if you already know your direction.
- 🧑 Actor profiling — builds a structured model of WHO you are: skills, resources, constraints, values, motivations. Not a form — a dialogue.
- 🌍 Landscape reconnaissance — searches 150+ web results and 80+ papers to map the terrain of your chosen field. Evaluates maturity, competition, entry barriers.
- 🎯 Direction narrowing — presents ranked sub-directions within your field, matched to your specific profile. You choose; it doesn't choose for you.
- ⚔️ Obstacle analysis — identifies barriers (knowledge, resource, capability, competition), assesses severity, proposes evidence-backed mitigations. If obstacles are unacceptable, backtracks to alternative directions.
- 🌳 Goal decomposition — KAOS-style AND/OR tree decomposition of your top goal into actionable, verifiable sub-goals. Feasibility-checked against your profile.
- 💎 North Star synthesis — crystallizes everything into one sentence: "[verb] [goal], through [path], solving [problem], ultimately [impact]"
- 📋 Research Brief generation — aggregates all context into a structured document for downstream research strategies (e.g., DARE's intake)
Because most people don't know. And those who think they know often haven't validated their direction against reality — the field might be saturated, the barriers might be insurmountable, or a better direction might be hiding one step away.
North Star Crystallization treats research direction as a requirements engineering problem. The user is a stakeholder with fuzzy needs; the skill's job is to elicit, structure, validate, and crystallize those needs into actionable requirements.
This is not a pipeline. It's a strategy book — CC reads it and makes autonomous decisions about:
- Which tactics to deploy and in what order
- How deep to go (cold-start goes deep; hot-start skips most)
- When to backtrack (obstacles rejected → return to direction selection)
- When to iterate (goal decomposition fails validation → refine)
The human provides direction and decisions. The AI provides structure and investigation.
Strategy (cold-start | warm-start | hot-start)
→ War doctrine: WHAT to accomplish, HOW MUCH depth
→ Available tactics + execution order guidance
Tactic (6 available)
→ Methodology: HOW to combine SOPs into coherent workflows
→ Available SOPs + sequencing logic
SOP (23 total: 11 dialogue + 10 subagent + 2 import)
→ Execution: HOW to do one specific thing
→ Either talks to user, spawns an agent, or imports external skill
Each layer has a single concern. A Strategy never executes an SOP directly. A Tactic never decides research direction. Strict layering = independently testable, replaceable, composable.
| Framework | What It Contributes | Where It Appears |
|---|---|---|
| KAOS (van Lamsweerde) | Goal → Sub-goal → Obstacle → Resolution | goal-decomposition tactic |
| i* (Yu) | Actor Modeling + Intentionality (WHY questions) | actor-profiling tactic |
| NFR Framework | Softgoal decomposition for feasibility | feasibility-check SOP |
| Requirements Engineering | Elicitation → Analysis → Specification → Validation | Overall flow structure |
┌───────────────────────────────────────────────────────────────┐
│ ENTRY POINT (ENTRY.md) │
│ Routes to strategy based on user's information density │
├───────────────────────────────────────────────────────────────┤
│ STRATEGY (3) │
│ cold-start, warm-start, hot-start │
├───────────────────────────────────────────────────────────────┤
│ TACTIC (6) │
│ actor-profiling, landscape-reconnaissance, │
│ direction-narrowing, obstacle-analysis, │
│ goal-decomposition, north-star-synthesis │
├───────────────────────────────────────────────────────────────┤
│ SOP (23) │
│ dialogue (11) | subagent (9) | import (3) │
├───────────────────────────────────────────────────────────────┤
│ EXTERNAL SKILLS (MCP servers — atomic operations) │
│ brave-search, alphaxiv, semantic-scholar, │
│ web-browsing, literature-engine, subagent-spawning │
└───────────────────────────────────────────────────────────────┘
All skills are flat under skills/ — no nested strategy/tactic/sop/ subdirectories. This means users can cp -r skills/ ~/.claude/skills/ and immediately have all 33 skills available as / commands.
north-star-crystallization/
├── ENTRY.md # Root entry point
├── skills/
│ ├── north-star-crystallization/ # Campaign SKILL.md
│ │
│ │ # Strategies (3)
│ ├── cold-start/ # Full flow for zero-context users
│ ├── warm-start/ # Partial flow for users with some direction
│ ├── hot-start/ # Minimal flow for users with clear direction
│ │
│ │ # Tactics (6)
│ ├── actor-profiling/ # WHO is the researcher
│ ├── landscape-reconnaissance/ # WHAT does the field look like
│ ├── direction-narrowing/ # WHERE specifically to go
│ ├── obstacle-analysis/ # WHAT stands in the way
│ ├── goal-decomposition/ # HOW to break it down
│ ├── north-star-synthesis/ # CRYSTALLIZE the final output
│ │
│ │ # SOPs — Dialogue (11)
│ ├── explore-resume/ # Understand background
│ ├── clarify-resources/ # Map available resources
│ ├── ask-constraints/ # Surface constraints
│ ├── ask-intentionality/ # Uncover motivations
│ ├── present-and-ask/ # Show panorama, gather preferences
│ ├── present-candidates/ # Show ranked sub-directions
│ ├── ask-obstacle-acceptance/ # User accepts or rejects obstacles
│ ├── formulate-top-goal/ # Formal goal statement
│ ├── ask-decomposition-validation/ # User confirms goal tree
│ ├── crystallize-north-star/ # Final one-sentence output
│ ├── final-validation/ # Quality gate + user confirm
│ │
│ │ # SOPs — Subagent (10)
│ ├── generate-candidate-fields/ # Generate field candidates
│ ├── landscape-synthesis/ # Evaluate fields → FieldPanorama
│ ├── deep-web-search/ # Full-page web reading (isolated context)
│ ├── identify-obstacles/ # Enumerate barriers
│ ├── assess-obstacle-severity/ # Rate obstacle difficulty
│ ├── propose-mitigations/ # Evidence-backed solutions
│ ├── and-or-decompose/ # KAOS goal tree
│ ├── validate-leaves/ # Leaf node quality check
│ ├── feasibility-check/ # Reality check vs profile
│ ├── generate-research-brief/ # Aggregate all context
│ │
│ │ # SOPs — Import (2)
│ ├── broad-web-search/ # → web-browsing/web-search
│ └── broad-paper-search/ # → literature-engine/literature-overview
├── tests/
│ └── integration-prompt.md # Live integration test scenarios
└── README.md
Users can copy the entire skills/ directory into their .claude/skills/ to register all 33 skills as / commands.
| Strategy | Trigger | Depth |
|---|---|---|
| cold-start | User has no direction, no field, no constraints | Full: all 6 tactics, all SOPs, maximum search depth |
| warm-start | User has a field or vague direction | Moderate: skip/simplify actor-profiling, full landscape + obstacles |
| hot-start | User has specific topic + constraints | Minimal: quick profiling, skip landscape, focus on decomposition |
| Tactic | Purpose | Key SOPs |
|---|---|---|
| actor-profiling | Build structured model of the researcher | explore-resume, clarify-resources, ask-constraints, ask-intentionality |
| landscape-reconnaissance | Map the terrain of candidate fields | broad-web-search, deep-web-search, broad-paper-search, landscape-synthesis, present-and-ask |
| direction-narrowing | Select specific sub-direction | present-candidates + generate-candidate-fields |
| obstacle-analysis | Identify and mitigate barriers | identify-obstacles, assess-obstacle-severity, propose-mitigations, ask-obstacle-acceptance |
| goal-decomposition | Break goal into actionable sub-goals | formulate-top-goal, and-or-decompose, validate-leaves, feasibility-check, ask-decomposition-validation |
| north-star-synthesis | Produce final output artifacts | crystallize-north-star, generate-research-brief, final-validation |
| Type | Count | Execution Model |
|---|---|---|
| Dialogue | 11 | CC talks directly to user — one question at a time, multiple choice preferred |
| Subagent | 10 | CC spawns a focused sub-agent with a dedicated prompt.md |
| Import | 2 | Delegates to external skill (web-browsing or literature-engine) |
One sentence in the format:
"[verb] [specific goal], through [method/path], solving [what problem], ultimately [what impact]"
Quality criteria: specific (not vague), ambitious (worth a top venue), achievable (within user's capabilities + mitigations).
Structured context document with 8 sections:
- North Star — the one-sentence direction
- Actor Profile — who is doing this research
- Field Context — selected field + landscape evaluation
- Chosen Direction — specific sub-direction + rationale
- Obstacle Landscape — known barriers + mitigations + acceptance
- Goal Decomposition — GoalTree with priorities
- Key References — papers and resources discovered during the process
- Key Terms — vocabulary and concepts central to this research area
The ResearchBrief is a requirement context document — it tells downstream strategies WHAT to research and WHY, not HOW.
| Dependency | Repository | What It Provides |
|---|---|---|
| web-browsing | NOESYNTH/web-browsing | web-search (quick search) + web-research (deep page reading) |
| literature-engine | NOESYNTH/literature-engine | literature-overview + literature-search + literature-research |
| subagent-spawning | NOESYNTH/subagent-spawning | Subagent dispatch conventions for 9 subagent SOPs |
| Server | Purpose |
|---|---|
| brave-search | Web search API (used by web-browsing skills) |
| alphaxiv | Paper search, content extraction, PDF queries (arXiv) |
| semantic-scholar | Paper lookup, citations, references, recommendations |
Invoke the /north-star skill in a Claude Code session:
/north-star-crystallization
Or read the entry point directly:
Read ENTRY.md. I want to publish a paper but I have no idea what to work on.
- Claude Code with MCP servers configured (brave-search, alphaxiv, semantic-scholar)
- Sibling repos available: web-browsing, literature-engine, subagent-spawning
- Subagent support enabled (for subagent SOPs to spawn agents)
Part of the Yogsoth AI ecosystem. Built by Pthahnix.