Installs into .claude/skills of the current project.
Are you the author of Mdp Modeler?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/anubhavg-icpl-mdp-modeler)
---
name: mdp-modeler
description: Given a task description, produce a Markov Decision Process spec and flag formulation risks before training. Use when you need help with mdp modeler.
license: CC-BY-NC-SA-4.0
phase: 9
lesson: 1
metadata:
version: 1.0.0
tags: [rl, mdp, modeling]
---
Given a task (control / game / recommendation / LLM fine-tuning), output:
1. State. Exact feature vector or tensor spec. Justify Markov property.
2. Action. Discrete set or continuous range. Dimensionality.
3. Transition. Deterministic, stochastic-with-known-model, or sample-only.
4. Reward. Function and source. Sparse vs shaped. Terminal vs per-step.
5. Discount. Value and horizon justification.
Refuse to ship any MDP where the state is non-Markovian without explicit mention of frame-stacking or recurrent state. Refuse any reward that was not defined in terms of the target outcome. Flag any `γ ≥ 1.0` on an infinite-horizon task. Flag any reward range >100x the typical step reward as a likely gradient-explosion source.
