Developer Builds Shell Script to Auto-Scout and Score GitHub Repos Each Morning

·3 views

A developer has created a shell script called github-scout.sh that automatically searches GitHub every morning and scores open-source projects and Claude Code skills using the Claude AI API. The tool separates the free crawling phase from the paid scoring phase, capping candidates at 30 to keep token costs predictable and fixed. Discovered skills are automatically routed either to an auto-enablement folder or a quarantine directory depending on whether they contain potentially risky bash or script files. All output is written only to the user's local ~/.claude/ directory, deliberately avoiding iCloud-synced or system-protected areas. The script was built to solve a personal productivity problem: manually browsing GitHub inconsistently led to missed tools, repeated discoveries, and wasted effort rewriting already-existing utilities.

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 Releases CLI Tool to Objectively Score Claude AI Skill Files

A developer has built and released 'skilleval', a command-line tool designed to objectively measure how well Claude AI SKILL.md instruction files perform. The tool runs blind A/B tests by comparing Claude's outputs with and without a skill injected, using a randomized Gemini Flash judge to eliminate position bias. It assigns margin-based scores, tracks run history, and flags regressions when model updates or skill edits silently degrade performance. Teams can also compare two skill versions head-to-head or integrate the tool into CI pipelines to block pull requests that break existing skills. The project addresses what the developer calls 'vibes-based evaluation', where engineers informally judge skill quality without a shared, repeatable rubric.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer Builds Chess-Concept Tool to Classify Everyday Life Decisions

A developer has created Position Evaluator, a daily decision-logging app that maps personal dilemmas to one of seven chess concepts such as Zugzwang, Fork, and Gambit. The tool accepts a short description of a situation and uses Google's Gemini AI to classify it, returning a category, a confidence level, and a brief explanation. Built with Python Flask, MySQL, and a vanilla JS frontend, the app also tracks patterns over time via a dashboard showing which chess condition a user encounters most frequently. To ensure consistent classifications, the developer gave Gemini a strict priority order for evaluating concepts rather than letting it freely choose from a flat list. The project was submitted as part of DEV Community's Weekend Challenge: Passion Edition.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer Builds LLM Memory System That Tracks How Quickly Facts Go Outdated

A developer has created VoltMem, a memory layer for large language model agents designed to distinguish between stable and volatile facts. The system addresses a core flaw in existing LLM memory tools, where all stored facts are treated equally regardless of how quickly they become outdated. VoltMem assigns domain-specific volatility scores to memories — for example, a user's location is flagged as volatile while personality traits are treated as stable — adjusting how easily each fact can be overwritten. At retrieval time, the system also down-ranks stale volatile memories even when they remain semantically relevant to a query. The project draws on continual-learning research around the stability-plasticity tradeoff and has been benchmarked against the open-source memory tool Mem0 across several real-world scenarios.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer builds open-source HTML-to-DOCX converter using AI self-improvement loops

A developer frustrated with slow, error-prone backend Word document generation built a new open-source library called dom-docx to improve HTML-to-DOCX conversion fidelity. The project used 'Autoresearch loops,' an AI-driven technique where an agent autonomously iterates and self-improves against an objective scoring metric. The scoring system evaluated each conversion across 37 real-world HTML test cases, weighing visual layout fidelity at 50%, document editability at 35%, and compile speed at 15%. Benchmarked against established npm HTML-to-DOCX libraries, dom-docx achieved average layout-fidelity scores in the mid-90s percent compared to the mid-60s percent for competing tools. The library also produced schema-valid OOXML across all 37 test cases, with the largest performance gains seen in complex elements like nested lists, table backgrounds, and flex layouts.

0 comments Read more at DEV Community