OpenParable
Skip to content
← Back to skills

Prove

Testing & Quality
prove

Prove is an Agent Skill for testing & quality published in the Z3Prover/z3 repository on GitHub: a SKILL.md instruction file an AI agent can load to do this task.

Prove validity of logical statements by negation and satisfiability checking. If the negation is unsatisfiable, the original statement is valid. Otherwise a counterexample is returned.

View on GitHub ↗
QUICK START

How to use this skill

Bring this guide into your coding agent with a prompt tailored to the tool you use.

  1. Open your project in Codex.
  2. Copy and paste the prompt below.
  3. Review files and risks before approving any action.
Prompt to paste
Inspect the Agent Skill at https://github.com/Z3Prover/z3/blob/HEAD/.github/skills/prove/SKILL.md and all supporting files. Report provenance, permissions, scripts, dependencies, and security risks. Propose an exact installation into .agents/skills/prove, but wait for my approval before writing files or running scripts.

Copying this prompt does not install or run the skill. Review third-party files before use.

Given a conjecture (an SMT-LIB2 assertion or a natural language claim), determine whether it holds universally. The method is standard: negate the conjecture and check satisfiability. If the negation is unsatisfiable, the original is valid. If satisfiable, the model is a counterexample.

Step 1: Prepare the negated formula

Action: Wrap the conjecture in (assert (not ...)) and append (check-sat)(get-model).

Expectation: A complete SMT-LIB2 formula that negates the original conjecture with all variables declared.

Result: If the negation is well-formed, proceed to Step 2. If the conjecture is natural language, run encode first.

Example: to prove that (> x 3) implies (> x 1):

(declare-const x Int)
(assert (not (=> (> x 3) (> x 1))))
(check-sat)
(get-model)

Step 2: Run the prover

Action: Invoke prove.py with the conjecture and variable declarations.

Expectation: The script prints valid, invalid (with counterexample), unknown, or timeout. A run entry is logged to z3agent.db.

Result: On valid: proceed to explain if the user needs a summary. On invalid: report the counterexample directly. On unknown/timeout: try simplify first, or increase the timeout.

python3 scripts/prove.py --conjecture "(=> (> x 3) (> x 1))" --vars "x:Int"

For file input where the file contains the full negated formula:

python3 scripts/prove.py --file negated.smt2

With debug tracing:

python3 scripts/prove.py --conjecture "(=> (> x 3) (> x 1))" --vars "x:Int" --debug

Step 3: Interpret the output

Action: Read the prover output to determine validity of the conjecture.

Expectation: One of valid, invalid (with counterexample), unknown, or timeout.

Result: On valid: the conjecture holds universally. On invalid: the model shows a concrete counterexample. On unknown/timeout: the conjecture may require auxiliary lemmas or induction.

Parameters

ParameterTypeRequiredDefaultDescription
conjecturestringnothe assertion to prove (without negation)
varsstringnovariable declarations as "name:sort" pairs, comma-separated
filepathno.smt2 file with the negated formula
timeoutintno30seconds
z3pathnoautopath to z3 binary
debugflagnooffverbose tracing
dbpathno.z3-agent/z3agent.dblogging database

Either conjecture (with vars) or file must be provided.