Developer Builds File-Based Gate System to Keep AI Coding Agents on Track

·1 views

A developer has released an open-source project called sdlc-playbooks to address a common failure in AI-assisted coding: agents producing technically correct code that solves the wrong problem. The tool defines a seven-phase software delivery workflow — from design system and requirements through to a finished product — where each phase is governed by a playbook file rather than a persona-based agent prompt. Automated gate scripts enforce preconditions at each phase boundary, exiting with an error if required artifacts are missing or inconsistent, so rules cannot simply be ignored by the model. Human approval is built into the workflow as a mandatory checkpoint before implementation begins, keeping design and scope decisions with people rather than the agent. The project runs on both Claude Code and OpenAI Codex from a single source, and all workflow state is stored in files so the process can continue without the agent if needed.

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 1950s–60s Engineers Laid the Groundwork for Modern Reinforcement Learning

In 1959, Arthur Samuel tackled the impossibility of storing 500 billion checkers positions by replacing lookup tables with an adjustable scoring function trained through self-play, pioneering Temporal Difference learning — the idea of updating predictions continuously rather than waiting for a game to end. Donald Michie independently demonstrated policy-based learning in 1961 using a physical matchbox system called MENACE, establishing a second foundational branch of reinforcement learning alongside Samuel's value-based approach. Marvin Minsky formally named the credit-assignment problem in 1961, identifying why it is so difficult for machines to determine which past actions deserve credit for a later outcome. Michie's 1968 BOXES system took another leap by balancing a physical pole using only failure as feedback, eliminating the need for a human teacher entirely. These breakthroughs collectively addressed three core obstacles — memory limits, the absence of a world model, and reward hacking — and their solutions underpin virtually every reinforcement learning algorithm in use today.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Software Engineer Reflects on How AI Tools May Be Eroding Independent Thinking

A software engineer writing on DEV Community describes how daily use of AI tools like ChatGPT, Gemini, Claude, and Copilot has dramatically sped up tasks ranging from coding to home project planning. While he credits AI with making him significantly more productive, he notices a growing habit of outsourcing not just information retrieval but also analysis and decision-making to these tools. He contrasts this with pre-AI research habits, where sifting through multiple sources, spotting contradictions, and forming independent conclusions built critical thinking skills. The convenience of instant, well-organized AI answers, he argues, may be quietly replacing the effortful mental work that deeper understanding requires. His reflection raises a broader question about whether AI efficiency gains come at the cost of cognitive independence.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Indie Dev Shares Hard-Won Lessons on Visual Cohesion From Two Godot Games

A solo indie developer building two Android games in Godot — a storybook solitaire called Glimmerbook and an action RPG called Ithaca Rising — found that functional gameplay alone did not make either title feel polished. Both games suffered from visual inconsistency, where individually acceptable assets clashed with each other in layout, scale, and color. For Ithaca Rising, the team settled on a unified bronze-and-navy pixel UI family across all HUD and control elements to bring coherence to a busy screen. Glimmerbook received illustrated card surfaces, a burgundy woven card back, and story-themed icons to replace generic interface elements. The developer concluded that polish is not about the number of assets but about the visual agreement between assets, layout, motion, and gameplay.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Gas Audit Flags Up to $1.8M Daily Savings and DoS Risks in Maple Finance

A gas optimization audit of Maple Finance, an institutional lending protocol managing approximately $3 billion in total value locked on Ethereum and several Layer 2 networks, was conducted on 25 September 2026. The audit identified six key inefficiency areas across core contracts including PoolFactory, Loan, CreditLine, and Staking modules, with per-transaction gas waste ranging from 2,000 to 22,000 gas units. Implementing the top five recommendations could reduce end-user transaction fees by an estimated $1.2 million to $1.8 million daily, based on a 30-day window at 30 gwei on Ethereum mainnet. Beyond cost inefficiencies, auditors flagged exploitable attack vectors including out-of-gas denial-of-service attacks via inflated reward-claim arrays and front-running opportunities enabled by high-cost borrow functions. The overall codebase was assessed as well-architected, with most inefficiencies attributed to legacy Solidity patterns and the absence of batch processing for reward distributions.

0 comments Read more at DEV Community