Why AI Agent Pull Requests Fail and How Repo Contracts Can Fix It

·8 views

Developer Bobai Kato published an article on DEV Community on July 16 exploring why pull requests submitted by AI agents frequently get rejected. The piece examines the concept of 'repo contracts' as a structured solution to improve AI agent reliability in code repositories. Kato argues that clear repository-level agreements or specifications can help AI agents better understand project standards before submitting changes. The article is tagged under topics including AI agents, repo readiness, verification, and agent safety, and is estimated to be a five-minute read.

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)

How Pleadwise Uses Deterministic Code Over AI to Format Court-Ready Legal Filings

Developer building Pleadwise, a court-document tool for self-represented litigants, found that legal tech requires a fundamentally different approach than typical AI-driven products. Unlike creative tools where variation is acceptable, court filings must meet strict, exact formatting rules on margins, captions, line spacing, and page numbering. The solution involved structuring filings as validated data objects and storing each court's formatting requirements as separate, source-linked configuration profiles. This architecture allows formatting rules to be updated independently without rebuilding the entire document workflow. The key principle applied was to reserve AI for tasks where variation adds value, while using deterministic software for precision-critical legal requirements.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How to Build a Python MCP Nutrition Server Using DietlyAPI and Claude

A developer tutorial on DEV Community outlines how to build a Model Context Protocol (MCP) server in Python that gives AI clients like Claude and Cursor access to real-time nutrition data. The server wraps DietlyAPI and exposes two focused tools — search_foods and get_food — to retrieve accurate macro information on demand rather than relying on a language model's often-unreliable memory. The guide uses Python's FastMCP helper alongside the httpx library, with each tool implemented as an async function that queries DietlyAPI endpoints and returns raw JSON. The server communicates over stdio, requires no open ports for local use, and can be connected to AI hosts like Claude Desktop by adding a simple configuration entry. The tutorial emphasizes keeping tools narrow and preserving uncertainty in the data, so AI models report sourced values instead of generating plausible but incorrect nutritional figures.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How Dietly Made ChatGPT Its Biggest Traffic Source Through Clear, Cited Content

Nutrition platform Dietly found that AI assistants like ChatGPT became its top referral source, surpassing the generic search traffic it had anticipated at launch. The team discovered that pages answering questions directly in the opening paragraph, with concrete figures and named qualifications, were far more likely to be cited by AI tools than broad, slow-to-the-point articles. Editorial discipline played a key role: disputable nutrition claims were linked to health bodies or primary research, and data limitations — such as reliance on Open Food Facts — were stated openly rather than obscured by vague terms like 'verified.' Dietly also adopted structural practices including semantic headings, facts tables, visible review dates, and an llms.txt file to help AI systems identify useful starting points across the site. The team cautioned against treating AI referrals as guaranteed, noting that retrieval behavior changes frequently and that sustainable results come from continuously improving individual pages rather than mass-producing near-identical content.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How Dietly Serves 850,000 Food Pages From a Single Server Using Static HTML

Dietly, a food nutrition platform, pre-renders over 850,000 food pages as static HTML files during build time rather than generating them dynamically on each request. This approach eliminates database and API calls from the reader path, with Caddy and Cloudflare handling delivery of the pre-built files. Each food page is assigned a stable, sharded URL and tracked via a content hash and sitemap timestamp, so incremental builds only rewrite pages whose data has actually changed. Dynamic features like search and barcode lookups remain API-driven, while the static layer handles the bulk of public page traffic. The architecture keeps the generated web root on a separate volume, ensuring clean operational boundaries and preventing accidental 404s during rebuilds.

0 comments Read more at DEV Community