Developer releases open-source tool to catch AI agent breakage after model updates

·1 views

A developer has released an open-source Python library called toolcontract to detect when AI model updates silently change an agent's tool-calling behavior. The library works by pinning a set of expected tool calls as a contract and re-running them against the live model to check for structural differences. Unlike evaluation frameworks such as promptfoo or DeepEval, toolcontract focuses narrowly on whether a tool call's structure, arguments, and trajectory remain identical after a provider ships a new model version. It supports OpenAI, Anthropic, and LiteLLM-compatible models, and integrates with pytest or a standalone CLI. The project is available on PyPI and GitHub under an MIT license at version 0.1.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Developer Turns Discord Bot Experiment into Full AIOps Platform AcornOps

A developer maintaining the open-source project React ChatBotify began experimenting in mid-2025 with a Discord bot that could help troubleshoot infrastructure through conversation. The idea was shelved until January 2026, when a design thinking course reignited it through discussions with a collaborator named Ning. Together, they reimagined the concept beyond a simple chatbot, asking what a self-hosted, full-fledged AIOps platform would actually require. Months of iterative design and problem-solving led to the creation of AcornOps, a platform aimed at enabling AI to safely participate in operational tasks. The project is still evolving, with the team planning to share deeper details about its architecture, decisions, and lessons in a forthcoming article.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer Dives Deep into Spring Framework During Week 9 of 100DaysOfCode

A developer participating in the #100DaysOfCode challenge dedicated Week 9 to deepening their understanding of the Spring Framework rather than building new features. The week focused on core backend concepts including Spring AOP, JDBC, Caching, Reactive Programming, and Concurrency. A key insight gained was how Spring uses Aspect-Oriented Programming and proxy patterns to apply cross-cutting concerns like caching without embedding that logic directly in business methods. The developer also explored reactive design patterns, examining why non-blocking architecture matters for building scalable, modern applications. Alongside technical learning, they continued applying for backend engineering roles and preparing for technical interviews.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Raku Programming Language Handles Infinite Sequences With Lazy Evaluation

Raku, a programming language, natively supports infinite sequences using the infinity symbol (∞) and lazy evaluation, computing values only when explicitly requested. Developers can define sequences like the Fibonacci series with a simple pattern rule and retrieve any element on demand without pre-calculating the entire list. The language also allows mapping and filtering operations on infinite sequences, such as squaring numbers or extracting prime values, using built-in methods like map and grep. A laziness check via the is-lazy method confirms that such sequences are evaluated on demand rather than upfront. This is the first part of a series exploring Raku's capabilities with infinite and lazy data structures.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer Builds Blockchain Tool to Verify Dog Walks Using GPS Hashing on Solana

A developer created 'Proof of Walk,' an app that cryptographically verifies dog walking routes using Solana's devnet blockchain. The walker's phone records a GPS route, which is hashed in the browser and committed as a signed attestation on-chain, while the raw location data never leaves the device. Pet owners can later verify the walk by comparing a locally recomputed hash against the on-chain record, making route tampering detectable. Even shifting a single GPS coordinate by roughly 11 metres produces a completely different hash, instantly flagging any alteration. The developer acknowledges the tool narrows but does not eliminate the trust gap, as it proves a device recorded a route rather than confirming a dog was actually present.

0 comments Read more at DEV Community