Full Stack HQ Brings Approval-Gated Control Layer to AI Coding Agents

·59 views

Full Stack HQ is an open-source engineering framework designed to make AI coding agents more predictable by requiring explicit human approval before executing any planned changes. The project supports three major AI coding tools — Claude Code, Google Antigravity IDE, and OpenAI Codex — through dedicated host adapters and configuration files. Instead of a single general-purpose agent, the stack offers ten specialist AI personas covering areas such as frontend, backend, security, and DevOps, alongside 28 domain-specific knowledge modules. A core concern the project addresses is agents taking irreversible actions without user confirmation, so all workflows are structured into phased plans that pause and await a defined approval keyword before proceeding. The framework is installable via a GitHub repository and targets developers who want structured, auditable control over AI-assisted coding workflows.

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 Builds Idiocracy-Themed CMS Demo Using Sanity Workflows and AI Agent

A developer has built a satirical content management project inspired by the film Idiocracy, using Sanity CMS to simulate the fictional Department of Agriculture. In the demo, an AI agent powered by Claude drafts crop-watering proposals, which a human 'Cabinet' approves or rejects via a custom picture-button Studio interface. Sanity Workflows then drives the approved plan through watering and harvesting stages on a live Next.js front end. The project was built in a single Claude Code session and recorded a perfect score of 27 out of 27 timed edits across three interfaces. It was submitted as an entry for the Sanity Challenge and is available as an open-source repository on GitHub.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Five Practical Systems to Stay Productive, Visible, and Balanced While Working Remotely

Remote work offers flexibility but removes the natural boundaries that offices provide, making it easy for work to bleed into personal time. Experts and experienced remote workers recommend building start and end-of-day rituals to mentally separate work from personal life. Maintaining a private log of completed tasks and achievements helps employees advocate for themselves during performance and compensation reviews. Communicating asynchronously with complete, well-documented messages respects teammates' focus time and builds a reputation for professionalism. Setting a firm daily logout time and redirecting commute-related savings into investments are also highlighted as key habits for long-term career sustainability.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer Converts 2014 Samsung Galaxy Note 4 into a 1.2W Linux Marine Safety Server

A software developer has repurposed a decade-old Samsung Galaxy Note 4 into a low-power Linux server running on just 1.2 watts, using Termux and headless Linux installed on the 2014 device. The phone now powers Maritime Watch, an open-source coastal safety tool that ingests hourly marine weather data and live AIS vessel traffic to generate tailored departure advisories for small-boat fishermen. Unlike generic weather apps, the system evaluates conditions against three hull-size categories, sending subscribers a daily 6 AM notification with specific safe departure windows and return deadlines. The developer chose the retired smartphone over a Raspberry Pi or a paid cloud server, citing its built-in battery backup, integrated wireless connectivity, and zero hardware cost. Taming Android's aggressive power management required disabling background sleep killers via Termux wake-lock and exempting the app from the OS battery optimization settings.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How MCP Tools Can Replace Manual Hex Pasting in Crypto KAT Debugging

When a Known Answer Test fails on a cryptographic primitive, developers typically diagnose the issue by pasting raw hex and ACVP JSON into a chat prompt, but this approach breaks down with large vector files. For small-scale testing, running the KAT binary locally and copying only the failing vector ID and byte offset into a prompt remains a practical and auditable method. The article proposes exposing KAT runners as Model Context Protocol tools over stdio for cases involving massive ACVP payloads and repeated test cycles. Two key tools are suggested: one to run the crypto KAT and return structured status with a diff window, and another to load ACVP vectors with filtering instead of dumping entire files. The core principle is keeping tool outputs small and typed so the AI agent requests additional context on demand rather than receiving the full file by default.

0 comments Read more at DEV Community