CompSci Math Agent Tools

A Claude Code agent toolkit implementing the five core mathematical domains from MIT's "Mathematics for Computer Science" (Lehman, Leighton, Meyer).

Quick Reference

Domain	Primary Use Cases	Key Tools
Proofs	Logical verification, induction	`evaluate_proposition`, `verify_induction`, `check_proof`
Structures	Graph algorithms, cryptography	`gcd_extended`, `rsa_keygen`, `graph_analyze`
Counting	Complexity analysis, combinatorics	`asymptotic_compare`, `binomial`, `generating_func`
Probability	Risk assessment, distributions	`bayes_update`, `expected_value`, `chernoff_bound`
Recurrences	Algorithm complexity	`master_theorem`, `solve_recurrence`, `akra_bazzi`

Tool Invocation Pattern

All tools follow a consistent interface:

python

from compsci_math import MathAgent

agent = MathAgent()
result = agent.invoke("tool_name", **parameters)

Domain 1: Proofs (Chapters 0-7)

Core logical operations and proof verification. See references/PROOFS.md for full theory.

Proposition Evaluation

Evaluate compound logical statements with all standard connectives.

python

# Direct evaluation
agent.invoke("evaluate_proposition", 
    expression="(P AND Q) IMPLIES (P OR R)",
    assignments={"P": True, "Q": False, "R": True})

# Truth table generation
agent.invoke("truth_table", variables=["P", "Q"], expression="P IMPLIES Q")

Induction Verification

Verify induction proofs with base case and inductive step checking.

python

agent.invoke("verify_induction",
    property="sum(1..n) = n*(n+1)/2",
    base_case=1,
    domain="positive_integers")

Predicate Quantifiers

Evaluate universal (∀) and existential (∃) quantifiers over finite domains.

python

agent.invoke("for_all", predicate="x > 0", domain=range(1, 100))
agent.invoke("exists", predicate="x % 7 == 0", domain=range(1, 50))

Domain 2: Structures (Chapters 8-12)

Number theory and graph structures. See references/STRUCTURES.md for full theory.

Number Theory

python

# Extended Euclidean Algorithm
agent.invoke("gcd_extended", a=102, b=70)  # Returns (gcd, s, t) where gcd = s*a + t*b

# Modular exponentiation
agent.invoke("mod_exp", base=3, exp=100, mod=17)

# Euler's totient
agent.invoke("euler_phi", n=60)

RSA Operations

python

# Key generation
keys = agent.invoke("rsa_keygen", bits=2048)

# Encryption/Decryption
cipher = agent.invoke("rsa_encrypt", message=42, public_key=keys["public"])
plain = agent.invoke("rsa_decrypt", cipher=cipher, private_key=keys["private"])

Graph Analysis

python

# Build and analyze graphs
agent.invoke("graph_create", vertices=5, edges=[(0,1), (1,2), (2,3), (3,4), (4,0)])
agent.invoke("topological_sort", dag=graph)
agent.invoke("is_bipartite", graph=graph)
agent.invoke("chromatic_number", graph=graph)

Domain 3: Counting (Chapters 13-15)

Asymptotic analysis and combinatorics. See references/COUNTING.md for full theory.

Asymptotic Comparison

python

# Compare growth rates
agent.invoke("asymptotic_compare", f="n^2", g="n*log(n)")  # Returns "f = Ω(g)"

# Classify complexity
agent.invoke("complexity_class", expression="3n^2 + 5n*log(n) + 100")  # Returns "Θ(n²)"

Combinatorics

python

# Binomial coefficients
agent.invoke("binomial", n=52, k=5)  # C(52,5) = 2,598,960

# Multinomial
agent.invoke("multinomial", n=10, groups=[3, 3, 4])

# Permutations
agent.invoke("permutations", n=10, k=3)  # P(10,3) = 720

# Inclusion-Exclusion
agent.invoke("inclusion_exclusion", set_sizes=[100, 80, 60], intersections={...})

Generating Functions

python

# Extract coefficient
agent.invoke("gf_coefficient", 
    generating_function="x / (1 - x - x^2)",  # Fibonacci GF
    n=10)  # F_10 = 55

# Solve counting problem
agent.invoke("coin_change", denominations=[1, 5, 10, 25], target=100)

Domain 4: Probability (Chapters 16-20)

Probability theory and distributions. See references/PROBABILITY.md for full theory.

Bayesian Inference

python

# Bayes' theorem
agent.invoke("bayes_update",
    prior=0.01,           # P(disease)
    likelihood=0.95,       # P(positive | disease)
    false_positive=0.05)   # P(positive | no disease)

Distributions

python

# Compute probabilities
agent.invoke("binomial_prob", n=10, k=3, p=0.5)
agent.invoke("poisson_prob", k=5, lambda_=3.2)
agent.invoke("geometric_prob", k=4, p=0.3)

Expectation and Variance

python

# Discrete random variable
agent.invoke("expected_value", distribution={(0, 0.2), (1, 0.5), (2, 0.3)})
agent.invoke("variance", distribution={(0, 0.2), (1, 0.5), (2, 0.3)})

Probability Bounds

python

# Markov's inequality
agent.invoke("markov_bound", expected_value=10, threshold=50)

# Chebyshev's inequality
agent.invoke("chebyshev_bound", variance=25, deviations=2)

# Chernoff bounds
agent.invoke("chernoff_bound", n=100, p=0.5, delta=0.1)

Domain 5: Recurrences (Chapter 21)

Recurrence relations and algorithm analysis. See references/RECURRENCES.md for full theory.

Master Theorem

python

# Apply Master Theorem: T(n) = aT(n/b) + f(n)
agent.invoke("master_theorem", a=2, b=2, f="n")
# Returns: "Case 2: T(n) = Θ(n log n)"

agent.invoke("master_theorem", a=4, b=2, f="n")
# Returns: "Case 1: T(n) = Θ(n²)"

Linear Recurrence Solver

python

# Solve homogeneous linear recurrence
agent.invoke("solve_linear_recurrence",
    coefficients=[1, 1],     # T(n) = T(n-1) + T(n-2)
    initial_values=[0, 1])   # T(0)=0, T(1)=1 (Fibonacci)
# Returns closed form: "(φⁿ - ψⁿ)/√5"

Akra-Bazzi Formula

python

# General divide-and-conquer
agent.invoke("akra_bazzi",
    terms=[(1, 0.5), (1, 0.33)],  # T(n/2) + T(n/3)
    f="n")
# Returns p ≈ 0.788 and asymptotic solution

Recurrence Evaluation

python

# Compute terms with memoization
agent.invoke("evaluate_recurrence",
    recurrence="T(n) = 2*T(n-1) + 1",
    base_cases={1: 1},
    n=20)

Implementation Files

The skill includes reference implementations in both Cyber and Zig:

File	Language	Description
`tools/math_proofs.py`	Python	Proof verification tools
`tools/math_structures.py`	Python	Number theory and graphs
`tools/math_counting.py`	Python	Combinatorics and asymptotics
`tools/math_probability.py`	Python	Probability computations
`tools/math_recurrences.py`	Python	Recurrence solvers

Agent Integration

For Claude Code integration, import the unified agent interface:

python

from tools.math_agent import CompSciMathAgent

agent = CompSciMathAgent()

# Route to appropriate domain automatically
result = agent.query("What is the complexity of T(n) = 2T(n/2) + n?")
result = agent.query("Compute C(52, 5)")
result = agent.query("Apply Bayes' theorem with prior 0.01 and likelihood 0.95")

Error Handling

All tools return structured results:

python

{
    "success": bool,
    "result": <computed_value>,
    "explanation": str,          # Human-readable reasoning
    "latex": str,                # LaTeX representation
    "complexity": str,           # Time/space complexity of computation
    "references": [str]          # Relevant theorem citations
}

Grace Hopper Integration

This toolkit supports the Praescientia prediction system's state checkpoint and rollback architecture. Mathematical computations produce immutable, hashable results suitable for blockchain-based consensus:

python

from tools.math_agent import CompSciMathAgent

agent = CompSciMathAgent(checkpointing=True)
result = agent.invoke("master_theorem", a=2, b=2, f="n")

# Result includes hash for state verification
print(result.state_hash)  # SHA-256 of computation state

Zig Acceleration (Local Claude Code)

For local Claude Code installations with network access to ziglang.org, the skill supports compiled Zig binaries for performance-critical operations. The Zig implementations offer significant speed advantages for large number operations, extensive iteration, and memory-intensive graph algorithms.

Prerequisites

Zig acceleration requires adding ziglang.org to your Claude Code network allowlist. See the implementation plan documentation for configuration details.

Installation

Run the Zig installation script to download and configure the compiler:

bash

./scripts/install_zig.sh

Using the Hybrid Agent

The HybridMathAgent automatically selects between Zig and Python backends based on availability:

python

from tools.zig_bridge import HybridMathAgent

agent = HybridMathAgent(prefer_zig=True)

# Uses Zig when available, falls back to Python otherwise
result = agent.invoke("gcd_extended", a=102, b=70)

# Check backend status
print(agent.get_backend_status())

Direct Zig Bridge Access

For fine-grained control over compilation and invocation:

python

from tools.zig_bridge import ZigBridge, ZigModule

bridge = ZigBridge()

# Check Zig availability
if bridge.is_zig_available():
    # Compile a specific module
    bridge.compile_module(ZigModule.STRUCTURES, optimize=True)
    
    # Invoke a function
    result = bridge.invoke(
        ZigModule.STRUCTURES, 
        "gcd_extended", 
        {"a": 102, "b": 70}
    )
    print(result.result)

Zig-Accelerated Tools

The following tools have Zig implementations available:

Tool	Module	Typical Speedup
`gcd`	STRUCTURES	5-10×
`gcd_extended`	STRUCTURES	5-10×
`mod_exp`	STRUCTURES	10-50×
`euler_phi`	STRUCTURES	5-10×
`is_prime`	STRUCTURES	10-100×
`binomial`	COUNTING	3-5×
`permutations`	COUNTING	3-5×
`evaluate_recurrence`	RECURRENCES	10-50×

Fallback Behavior

When Zig is unavailable (hosted Claude Code environment), all operations transparently fall back to the Python implementations with no change to the API. Code written for the hybrid agent works in both environments.