Developer builds infrastructure DSL with 5 compilation targets in 10 days using AI

·1 views

A software developer built 'Infra Lang', a domain-specific language (DSL) for infrastructure-as-code that compiles a single source file into five targets: Kubernetes manifests, Docker Compose, Helm charts, Terraform HCL, and GitHub Actions workflows. The project was completed in 10 days using an AI-assisted development process, where the developer retained all architectural decisions while using AI to write implementation code and tests. The DSL includes a hand-written LALR(1) parser, built-in security and reliability linting rules, and a Language Server Protocol (LSP) for editor support. Testing against real Kubernetes clusters and multiple operating systems uncovered 27 bugs that unit tests alone had missed. The developer noted that high line coverage proved misleading, with some modules scoring only 34% on mutation testing until targeted fixes raised critical modules to 82–100%.

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)

BrightbeamAI Proposes Open Protocol for Human-AI Agent Collaboration

BrightbeamAI has published an open-source specification called the Collaborative Human Agent Protocol, or CHAP, on GitHub. The protocol appears to define a standardized framework for interaction between humans and AI agents. It was shared on Hacker News, where it received minimal initial engagement with four points and no comments. The project is in early stages, with the repository serving as the primary source of documentation and technical details.

0 comments Read more at Hacker News

ProgrammingDEV Community ·

JPG to PNG Conversion: How Engineers Choose the Right Method for Any Batch

A software engineer has outlined a practical decision framework for converting JPEG images to PNG, drawing on experience across three real-world scenarios ranging from a 40,000-image CMS migration to a five-file Slack request. The guide highlights that switching formats is not always necessary, and teams should only convert when a downstream system requires PNG, transparency is needed, or a pipeline enforces a single container type. A key technical consideration is file size: a 200 KB JPEG can balloon to 1.4 MB as a PNG, impacting bandwidth costs and CDN budgets. For small batches under roughly 20 files, manual export via an image editor is recommended, while larger local batches are better handled with command-line tools like ImageMagick or Python's Pillow library. The core advice is that the right conversion method depends on batch size, privacy requirements, deadline pressure, and what the end consumer actually expects.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Google Gemini Adds Interactive 3D Visualizations in Chat, Excludes Workspace Users

Google has rolled out interactive 3D visualizations and charts within the Gemini app, enabling users to explore rotating models, fractals, and simulations directly inside a chat interface. Users can trigger the feature with prompts like 'show me' or 'help me visualize,' with some outputs including adjustable sliders to manipulate variables. The capability is designed to support scientific, technical, and educational use cases where visual exploration is more effective than text-based responses. The rollout is available globally but is restricted to the Pro model at gemini.google.com, excluding Education and Workspace account holders for now. Organizations considering integrating the feature into managed environments should note that access is not yet extended to enterprise or education accounts.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How a typo and a permission quirk sent an AI agent chasing phantom data for days

A small automated publishing operation running six scheduled AI agent sessions daily discovered that two days of performance metrics — showing four articles published and zero replies — were entirely wrong due to an unverified account handle in its queries. The platform's API responded with valid, empty arrays each time because the queried username simply did not exist, masking the actual record of five published articles and two reader comments. Separately, the team uncovered a selective URL permission gate in their tooling: the agent could fetch any URL that was a prefix of one returned by a tool, but not URLs derived by extending those paths, the opposite of how filesystem or OAuth scopes typically work. Both failures produced confident, well-formed outputs with no error signals, making them harder to detect than outright crashes. The incidents led the team to add a provenance column to their reachability checks, distinguishing between tool-supplied URLs and manually typed ones.

0 comments Read more at DEV Community