Developer Builds Lightweight Memory Plugin to Persist Lessons Across OpenCode Sessions

·1 views

A developer grew frustrated with repeatedly explaining the same project-specific conventions and preferences to each new OpenCode AI coding session. To solve this, they created opencode-plugin-memory, a small plugin that gives the active OpenCode agent four tools to save, update, delete, and retrieve project-relevant lessons. Knowledge is stored as local Markdown files within the project and a compact index is automatically injected into future session prompts. Unlike AGENTS.md, which holds mandatory team-wide rules, this memory layer is designed for narrower, session-earned insights that are worth preserving but not formalizing. The plugin deliberately avoids cloud services, vector databases, or background pipelines, keeping all state visible and local to the repository.

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)

Unicode Has No Superscript 'q': The Gaps Behind Plain-Text Formatting

A developer building a free superscript text converter discovered that Unicode does not contain a superscript letter 'q', nor a complete set of superscript uppercase or subscript characters. Unlike bold or italic styling in HTML, Unicode encodes each superscript character as a separate codepoint, with entries scattered across four different blocks added by different communities over decades. This means converting text to superscript requires a manually assembled lookup table rather than any algorithmic transformation. Many existing converters silently substitute lookalike Greek or mathematical symbols for missing characters, which can cause problems in screen readers, search engines, and certain fonts. The developer chose instead to flag unsupported characters explicitly, prioritising accuracy over the appearance of completeness.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Vercel Expands CLI With DNS, Domain, and Project Management Capabilities

Vercel has updated its command-line interface to support DNS record management, domain renewals, and project member operations, all with JSON output for easier scripting. Previously, these tasks required manual use of the Vercel dashboard or direct API calls with self-managed tokens. The update closes a key gap for developers running automated or agent-driven deployment pipelines, making CLI commands composable with tools like jq and easier to integrate into CI workflows. Vercel also launched a production tracing feature that continuously samples live traffic without requiring request reproduction, priced at $0.50 per million span units with no charge until a sampling rule is activated. Additionally, MiniMax AI models are available free via Vercel AI Gateway through September 6, after which integrations without fallback routing configured may encounter errors rather than a silent switch to paid tiers.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

What Is GMTI Radar Processing and How Does It Detect Moving Targets?

Ground Moving Target Indication (GMTI) is a radar processing capability designed to detect moving objects on the ground while filtering out stronger returns from stationary terrain, buildings, and vegetation. The core challenge lies in separating motion-related radar signals from background clutter, which can be far more powerful than the signal from an actual moving target. GMTI systems rely on Doppler information — changes in the received signal caused by relative motion — to distinguish moving objects from their surroundings. Because airborne GMTI radars are themselves in motion, the processing pipeline must account for the aircraft's position, velocity, altitude, and orientation at all times. Rather than a single algorithm, GMTI is best understood as a multi-stage software pipeline spanning radar acquisition, navigation alignment, clutter suppression, detection, geolocation, and target tracking.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How Repeated App Store Rejections Became an Automated Pre-Submission Checklist

App store reviews by Apple and Google follow rules that are partly written and partly learned through repeated submission outcomes, with no way to test locally before submitting. One developer team turned every rejection pattern and rule change into an explicit pre-submission check run on their side before the app reaches the store. Two common rejection categories — apps appearing unfinished and mismatches between privacy declarations and actual app behavior — were addressed by embedding automated guardrails directly into the product. Advice such as 'fill your sections before submitting' was converted into enforced checks, since automated verification is more reliable than optional guidance. The team draws a clear line between what can be mechanically verified and what still requires expert human judgment before submission.

0 comments Read more at DEV Community