Developer shares 5 hard-learned lessons from building on HMRC's Making Tax Digital API

·1 views

A developer building TapTax, a Making Tax Digital app for UK sole traders, documented five unexpected technical challenges encountered while integrating with HMRC's APIs. Unlike most APIs, HMRC uses content negotiation via Accept headers for versioning, with different endpoints running on entirely different version numbers simultaneously. HMRC also mandates dozens of fraud prevention headers on every API call, with required fields varying depending on whether the app is web or mobile-based. A key conceptual trap is that quarterly tax submissions are cumulative year-to-date figures rather than per-quarter deltas, a distinction that can silently corrupt financial data if missed. The developer published these findings as engineering notes to help other developers avoid costly debugging sessions when working with HMRC's systems.

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 hardens Kademlia DHT security and automates Neovim config in one week

A developer submitted a pull request to py-libp2p this week addressing a peer identity binding vulnerability in the Kademlia Distributed Hash Table implementation. The fix explicitly binds signed PeerRecords to their signer's identity, making the DHT more resistant to spoofing and stale data propagation. Separately, the developer automated the resolution of lazy-lock.json merge conflicts in their Neovim configuration, reducing friction when syncing the setup across multiple machines. The week also included 167 lines of new study notes spanning topics such as xv6, Category Theory, and the Rust Handbook. In total, the developer logged 10 commits and 204 lines added across open-source and personal projects.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer Overhauls dev-publish Tool with 21 Commits for Production-Grade Resilience

A developer spent the week hardening their open-source dev-publish tool, pushing 21 commits across two repositories with over 11,000 lines of code changed. The primary focus was making the publishing orchestrator resilient to mid-run failures, ensuring it can resume interrupted multi-platform publishing without duplicating posts. Improvements also included safer tag normalization, stricter cover image validation with immediate error feedback, and a cleaner configuration by removing unused options. Better API response validation and a typed error system were added specifically for the Dev.to integration. The update was rounded out with an architecture diagram embedded in the README to document the tool's overall design and re-run behavior.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Spanlens: Open-Source Platform to Monitor, Trace, and Optimize LLM API Calls

Spanlens is a free, MIT-licensed observability platform designed to help developers monitor API calls made to major LLM providers including OpenAI, Anthropic, Gemini, Mistral, and local Ollama models. Setup requires just one line of code, either by redirecting the client's base URL to the Spanlens proxy or running a CLI wizard that automatically updates existing code. The platform logs every request with details such as token counts, latency, cost, and full prompt and response data, while its dashboard provides cost breakdowns per request, model, and end user. Advanced features include agent workflow visualization, anomaly detection against a 7-day rolling baseline, and real-time alerts via Email, Slack, or Discord. Spanlens also includes a PII and prompt-injection scanner, a cost-saving recommendation engine, prompt versioning with A/B testing, and an LLM-as-judge evaluation framework for scoring model outputs.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer shares system for managing 10,000+ AI prompts like software assets

A developer writing for DEV Community has outlined a structured system for organizing over 10,000 AI prompts built across multiple projects, books, and client work. Rather than treating prompts as throwaway conversation inputs, the approach classifies them as reusable software assets stored in purpose-based folder hierarchies. Each prompt is documented with metadata including category, intended model, and expected output, while versioning allows comparison between iterations over time. Prompts are also separated into generic and project-specific collections to maximize reusability across different workflows. The system is maintained in Git repositories, giving prompts the same version control and documentation treatment typically reserved for source code.

0 comments Read more at DEV Community