| name | doncheli-reasoning |
| description | Apply structured reasoning models to analyze a decision or problem. Activate when user mentions "reason", "pre-mortem", "5 whys", "pareto", "think through", "analyze decision", "inversion", "first principles", "second order". |
Don Cheli: Structured Reasoning Engine
Available Models
| ID | Best For |
|---|
| pre-mortem | Imagining failure before it happens; surfacing hidden risks |
| 5-whys | Root cause analysis of a known problem |
| pareto | Identifying the 20% of causes driving 80% of impact |
| inversion | Asking "what would guarantee failure?" to find blockers |
| second-order | Mapping downstream consequences of a decision |
| first-principles | Decomposing assumptions down to fundamental truths |
| minimize-regret | Choosing the option you will regret least in 10 years |
| opportunity-cost | Evaluating what you give up by choosing this option |
| circle-of-competence | Assessing whether you have the knowledge to decide well |
| map-territory | Separating your mental model from reality |
| probabilistic | Assigning probabilities to outcomes; expected-value calc |
| reversibility | Classifying decisions as one-way vs. two-way doors |
| rlm-chain-of-thought | Step-by-step explicit reasoning (RLM: PrimeIntellect style) |
| rlm-decomposition | Breaking a complex problem into atomic sub-problems (RLM) |
| rlm-verification | Self-checking the reasoning chain for contradictions (RLM) |
Instructions
- Read the problem or decision description
- If the user did not specify a model, suggest the 1–2 best-fit models with a one-line rationale; wait for confirmation or proceed with the top suggestion
- Apply the selected model(s) rigorously, showing all reasoning steps
- End with a clear recommendation or insight
Output Format
## Reasoning: <problem>
**Model applied:** <id> — <why this model fits>
### Analysis
<step-by-step application of the model>
### Insight / Recommendation
…
### Caveats
…
Quality Gate
- Show all intermediate steps — no conclusions without visible reasoning
- If using probabilistic model, all probabilities must sum to 1 across mutually exclusive outcomes
- If using rlm-verification, the final step must explicitly confirm or reject the reasoning chain
Do not use this skill when
- The question requires group deliberation with roles — use doncheli-debate instead
- The task is a straightforward implementation — use doncheli-implement instead
