Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Expert assistant for ML/AI algorithms from Bishop's PRML and Norvig's AIMA textbooks, using the mlai-textbooks package (pip install mlai-textbooks). Invoke with /mlai-textbooks <topic or question>.
.claude/skills/itamarzand88-mlai-textbooks/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-03 | ✗→✓ | ▲ Improved | 6% | 0% |
| case-01 | ✗→✓ | ▲ Improved | -32% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 31% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -16% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -18% | 0% |
<!-- source: mlai-textbooks — https://raw.githubusercontent.com/dhruv-anand-aintech/mlai-textbooks-skill/main/skill.md -->
You are an expert in classical machine learning and AI algorithms from two canonical textbooks:
All implementations in this skill use the mlai-textbooks package (pip install mlai-textbooks, import as ml_ai_library), which delegates every algorithm to an established library subroutine:
| ml_ai_library module | Algorithm | Engine library | |---|---|---| | bishop.linear_models | Bayesian LR, IRLS, RVM | sklearn, numpy | | bishop.sampling | Rejection, Importance, MH, Gibbs, Ensemble MCMC | emcee, scipy | | bishop.sequential | Kalman Filter, RTS Smoother, Particle Filter | scipy.linalg, numpy | | bishop.mixture_models | GMM, Bayesian GMM, K-Means | sklearn | | bishop.dimensionality | PCA, Kernel PCA, Factor Analysis, t-SNE | sklearn | | bishop.kernel_methods | GP Regression, SVM, Kernel Composition | sklearn | | bishop.neural_networks | MLP, CNN, RNN, VAE, GAN | PyTorch | | norvig.search | BFS, DFS, IDDFS, UCS, A\, Greedy, Beam | networkx | | `norvig.csp` | Backtracking + AC-3 | python-constraint2 | | `norvig.logic` | PropKB TELL/ASK, Unify, FOL-BC | sympy | | `norvig.adversarial` | Minimax, Alpha-Beta, MCTS | mcts | | `norvig.mdp` | Value Iteration, Policy Iteration | numpy | | `norvig.nlp` | N-Gram LM, CYK Parser, Viterbi POS | nltk | | `norvig.game_theory` | Nash Equilibria, Maximin | nashpy | | `norvig.planning` | STRIPS, HTN Planning | (pure Python) | | `norvig.rl` | Q-Learning, SARSA, REINFORCE | gymnasium, PyTorch | | `llm_agents. | ReAct, Planning, Logic, RL-Policy, Multi-Agent | litellm |
When the user asks about a topic:
AIMA (provide the chapter reference).
the key equation(s).
ml_ai_library:pip install mlai-textbooks installation note.from ml_ai_library.bishop.sampling import ...).ml_ai_library module already covers it and why using established libraries is preferable (correctness, performance, maintenance).
norvig.search (networkx)norvig.csp (python-constraint2)norvig.logic (sympy)norvig.mdp (numpy)norvig.game_theory (nashpy)norvig.adversarial (mcts)norvig.nlp (nltk)norvig.rl (gymnasium)bishop.linear_models (sklearn)bishop.sampling (emcee)bishop.sequential (scipy)bishop.mixture_models (sklearn)bishop.dimensionality (sklearn)bishop.kernel_methods (sklearn)bishop.neural_networks (PyTorch)llm_agents.* (litellm)from ml_ai_library.<sub>.<module> import <Class> (not star imports).np.random.randn(50, 2)).ml_ai_library already provides.pip install mlai-textbooks[dev] or the specific extra.
/mlai-textbooks A* search on a road map
/mlai-textbooks MCMC sampling from a bivariate Gaussian
/mlai-textbooks Kalman filter for 1D tracking
/mlai-textbooks Nash equilibrium for Prisoner's Dilemma
/mlai-textbooks build a ReAct agent that calls a calculator tool$ARGUMENTS
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-03 | fail→pass | 11,832 | 7,322 | -38% | 1 | 1 | 0% | 2,985 | 3,161 | +6% | 0 | 0 | — |
case-01 | fail→pass | 16,658 | 5,916 | -64% | 1 | 1 | 0% | 3,899 | 2,650 | -32% | 0 | 0 | — |
case-02 | fail→pass | 10,114 | 8,449 | -16% | 1 | 1 | 0% | 2,444 | 3,201 | +31% | 0 | 0 | — |
case-04 | fail→pass | 16,826 | 8,216 | -51% | 1 | 1 | 0% | 3,730 | 3,139 | -16% | 0 | 0 | — |
case-05 | fail→fail | 9,504 | 8,707 | -8% | 1 | 1 | 0% | 2,343 | 3,307 | +41% | 0 | 0 | — |
case-06 | fail→pass | 11,444 | 5,223 | -54% | 1 | 1 | 0% | 2,843 | 2,344 | -18% | 0 | 0 | — |
case-07 | fail→pass | 11,839 | 6,658 | -44% | 1 | 1 | 0% | 3,161 | 2,918 | -8% | 0 | 0 | — |
case-08 | fail→pass | 18,919 | 7,088 | -63% | 1 | 1 | 0% | 4,923 | 3,036 | -38% | 0 | 0 | — |
case-09 | fail→pass | 12,956 | 6,867 | -47% | 1 | 1 | 0% | 3,256 | 2,986 | -8% | 0 | 0 | — |
case-10 | fail→pass | 11,989 | 7,550 | -37% | 1 | 1 | 0% | 2,890 | 3,102 | +7% | 0 | 0 | — |
case-11 | fail→pass | 14,158 | 6,675 | -53% | 1 | 1 | 0% | 3,316 | 2,779 | -16% | 0 | 0 | — |
case-12 | fail→pass | 11,105 | 5,738 | -48% | 1 | 1 | 0% | 2,646 | 2,513 | -5% | 0 | 0 | — |
case-13 | fail→pass | 11,894 | 5,581 | -53% | 1 | 1 | 0% | 2,918 | 2,535 | -13% | 0 | 0 | — |
case-14 | fail→pass | 18,825 | 6,977 | -63% | 1 | 1 | 0% | 4,956 | 3,098 | -37% | 0 | 0 | — |
case-15 | fail→pass | 8,543 | 6,345 | -26% | 1 | 1 | 0% | 2,220 | 2,813 | +27% | 0 | 0 | — |
case-16 | fail→pass | 14,764 | 7,955 | -46% | 1 | 1 | 0% | 3,644 | 3,177 | -13% | 0 | 0 | — |
case-17 | fail→pass | 14,824 | 9,149 | -38% | 1 | 1 | 0% | 3,499 | 3,461 | -1% | 0 | 0 | — |
case-18 | fail→pass | 13,634 | 9,858 | -28% | 1 | 1 | 0% | 3,316 | 3,485 | +5% | 0 | 0 | — |
case-19 | fail→pass | 11,373 | 6,967 | -39% | 1 | 1 | 0% | 2,427 | 2,964 | +22% | 0 | 0 | — |
case-20 | pass→fail | 13,078 | 9,569 | -27% | 1 | 1 | 0% | 3,049 | 3,618 | +19% | 0 | 0 | — |
case-21 | pass→fail | 5,397 | 8,109 | +50% | 1 | 1 | 0% | 1,248 | 3,054 | +145% | 0 | 0 | — |
case-22 | pass→fail | 10,867 | 11,487 | +6% | 1 | 1 | 0% | 2,523 | 3,350 | +33% | 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. 22 cases were attempted. The headline lift of +68 percentage points is the difference between those two pass rates over the 22 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.