Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Minimal crystallization strategy for users who already have a specific research topic or problem (e.g., "I want to improve CoT faithfulness in LLMs") and need structuring into a formal North Star. Heavily simplifies or skips exploration tactics, focusing on obstacle analysis, goal decomposition, and synthesis. Use when the user's first message reveals a specific, actionable research direction.
.claude/skills/yogsoth-ai-hot-start/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-06 | ✗→✓ | ▲ Improved | 106% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 64% | 0% |
| case-01 | ✗→✓ | ▲ Improved | 62% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 4% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -52% | 0% |
The user already knows their direction. Your job is to structure it, not explore alternatives.
All SOPs in this strategy follow these rules:
| Tactic | Purpose | |--------|---------| | actor-profiling | Understand who the user is | | landscape-reconnaissance | Broad, shallow field exploration | | direction-narrowing | Focus within chosen field(s) | | obstacle-analysis | Identify and mitigate barriers | | goal-decomposition | KAOS-style AND/OR goal structuring | | north-star-synthesis | Converge into North Star + ResearchBrief |
actor-profiling (heavily simplified) → landscape-reconnaissance (skipped or minimal)
→ direction-narrowing (heavily simplified) → obstacle-analysis (simplified)
→ goal-decomposition → north-star-synthesisThis is a reference, not a mandate. The user already knows their direction. landscape-reconnaissance and direction-narrowing may only need a few searches for context — or may be skipped entirely if the user's topic is already well-defined.
You are the general. This strategy gives you:
What you decide:
The only non-negotiable: the process ends with north-star-synthesis producing a North Star + ResearchBrief that the user confirms.
<!-- BEGIN available-tables (generated) -->
Optional, no fixed order; the final leaf is always a sop.
| Tactic | When to use | | --- | --- | | actor-profiling | Understand who the user is — background, resources, constraints, and deep motivations. Produces an ActorProfile that informs all downstream decisions. Use this tactic at the start of any crystallization process to build a model of the user's capabilities, limitations, and intent. | | direction-narrowing | Focus within the user's chosen field(s). Identify specific sub-directions through deep paper and web research, then present ranked candidates. Use after landscape-reconnaissance has identified fields of interest. | | goal-decomposition | Structure the user's chosen direction into a formal goal tree using KAOS-style AND/OR decomposition. Validate feasibility against ActorProfile and ObstacleReport. Use after obstacle-analysis confirms the direction is viable. | | landscape-reconnaissance | Broad, shallow exploration of candidate research fields. Understand what's out there before narrowing. Use when the user needs to discover which fields are available to them — especially in cold-start and warm-start scenarios. | | north-star-synthesis | Converge all accumulated context into a crystallized North Star statement and structured ResearchBrief. Performs self-review before presenting to user. Use as the final tactic in any start mode — this is where everything comes together. | | obstacle-analysis | Identify what blocks the user from pursuing their chosen direction, assess severity, propose mitigations with search-validated evidence, and get user acceptance. Use after direction-narrowing has identified a specific direction. |
<!-- END available-tables (generated) -->
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-06 | fail→pass | 20,934 | 6,658 | -68% | 1 | 1 | 0% | 950 | 1,956 | +106% | 0 | 0 | — |
case-07 | fail→pass | 9,014 | 7,772 | -14% | 1 | 1 | 0% | 1,319 | 2,163 | +64% | 0 | 0 | — |
case-08 | pass→pass | 15,507 | 9,807 | -37% | 1 | 1 | 0% | 2,085 | 2,481 | +19% | 0 | 0 | — |
case-01 | fail→pass | 7,627 | 11,656 | +53% | 1 | 1 | 0% | 1,238 | 2,011 | +62% | 0 | 0 | — |
case-02 | fail→pass | 12,349 | 5,330 | -57% | 1 | 1 | 0% | 1,805 | 1,882 | +4% | 0 | 0 | — |
case-03 | fail→pass | 25,927 | 5,141 | -80% | 1 | 1 | 0% | 3,669 | 1,761 | -52% | 0 | 0 | — |
case-04 | fail→pass | 9,698 | 6,581 | -32% | 1 | 1 | 0% | 1,462 | 2,026 | +39% | 0 | 0 | — |
case-05 | fail→fail | 3,049 | 8,002 | +162% | 1 | 1 | 0% | 390 | 2,155 | +453% | 0 | 0 | — |
case-09 | fail→pass | 9,620 | 8,668 | -10% | 1 | 1 | 0% | 1,358 | 2,175 | +60% | 0 | 0 | — |
case-10 | fail→pass | 22,715 | 6,301 | -72% | 1 | 1 | 0% | 3,362 | 1,990 | -41% | 0 | 0 | — |
case-11 | pass→pass | 16,472 | 7,777 | -53% | 1 | 1 | 0% | 2,301 | 2,207 | -4% | 0 | 0 | — |
case-12 | fail→fail | 21,702 | 6,924 | -68% | 1 | 1 | 0% | 2,993 | 1,975 | -34% | 0 | 0 | — |
case-13 | pass→pass | 21,337 | 7,309 | -66% | 1 | 1 | 0% | 2,989 | 2,131 | -29% | 0 | 0 | — |
case-14 | fail→fail | 17,052 | 6,420 | -62% | 1 | 1 | 0% | 2,836 | 2,080 | -27% | 0 | 0 | — |
case-15 | pass→fail | 21,792 | 7,630 | -65% | 1 | 1 | 0% | 3,043 | 2,106 | -31% | 0 | 0 | — |
case-16 | fail→fail | 4,637 | 8,152 | +76% | 1 | 1 | 0% | 693 | 2,229 | +222% | 0 | 0 | — |
case-17 | pass→pass | 18,295 | 8,957 | -51% | 1 | 1 | 0% | 2,490 | 2,242 | -10% | 0 | 0 | — |
case-18 | fail→pass | 12,078 | 7,256 | -40% | 1 | 1 | 0% | 1,728 | 2,163 | +25% | 0 | 0 | — |
case-19 | fail→fail | 9,923 | 6,786 | -32% | 1 | 1 | 0% | 1,446 | 1,987 | +37% | 0 | 0 | — |
case-20 | fail→pass | 8,129 | 5,235 | -36% | 1 | 1 | 0% | 1,223 | 1,810 | +48% | 0 | 0 | — |
case-21 | pass→pass | 20,375 | 11,736 | -42% | 1 | 1 | 0% | 2,888 | 2,996 | +4% | 0 | 0 | — |
case-22 | pass→fail | 25,193 | 7,419 | -71% | 1 | 1 | 0% | 4,824 | 2,071 | -57% | 0 | 0 | — |
case-23 | pass→fail | 9,429 | 3,588 | -62% | 1 | 1 | 0% | 1,550 | 1,583 | +2% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted. The headline lift of +30 percentage points is the difference between those two pass rates over the 23 comparable cases. 3 cases got worse with the skill loaded, and they are included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.