pstack Adds Meta-Skill That Generates Custom Formal Verification Skills

pstack introduced a meta-skill, /create-formal-verification-skill, that generates a customized verification skill and supporting suite based on the system property to be verified, producing TLA+-based plans for state machine and concurrency correctness and Kani-based plans for Rust undefined behavior. The skill is installable as a Grok Bot plugin, and the approach rests on formal verification proving invariants hold in all possible cases, with LLM agents taking over modeling and specification-implementation synchronization.
Event version 1.
Key insights
Meta-skill generates tailored verification suites
The /create-formal-verification-skill meta-skill does not perform formal verification directly; it generates a customized verification skill and supporting suite based on the system property to be verified. It produces TLA+-based plans for state machine and concurrency correctness and Kani-based plans for Rust undefined behavior.
Two technical routes for formal verification
The traditional route uses an external specification language such as TLA+ to model the system, then runs solvers or model checkers to exhaustively verify the specification, keeping code and proof separate. The newer route writes proofs in the same language, with Kani embedded directly in Rust and Bend expressing proofs as laws in its type system.
Historical bottlenecks cited for formal verification
Formal verification has remained theoretically attractive but outside mainstream practice due to two historical reasons: modeling is difficult, requiring specialized skill to abstract business requirements into formal models, and specifications are hard to keep synchronized with implementations.
LLM agents positioned to absorb modeling work
The two bottlenecks are identified as areas LLMs are suited to take over, with agents handling modeling and synchronization work. Formal verification provides agents strong feedback that traditional testing cannot, namely that a violation is mathematically impossible.