dsh-skill-recommender

You are running heavy daily workloads across Codex, Claude, and DSH, and the open-source skill ecosystem keeps adding new entries every week. Manually browsing awesome lists and GitHub search is slow, undirected, and frankly tedious. dsh-skill-recommender solves this with a purely local approach: it reads your session files from all three agent sources (no API keys required), distils a compact profile covering topic distribution, high-frequency tools, task types, and project directories, and uses that profile as the basis for every recommendation.

Armed with your profile, the plugin scans local skill directories, remote awesome lists, and even a full GitHub search (cached with a 6-hour TTL) to build a candidate pool. Each candidate is scored across four weighted dimensions - topic, tool, task, and proximity to what you already have installed - then gated by a tunable match index (0-100, default 60). Only skills that clear the threshold are returned, ranked top-N. The Web settings panel gives you real-time sliders for the match index, per-dimension weights, and skill-versus-plugin scope, plus per-dimension score breakdowns and direct GitHub links on every recommendation card.

This plugin is for you if you are already a daily DSH user, also work with Codex or Claude, and would rather get a curated short-list in seconds than spend half an hour scrolling GitHub. It pairs naturally with dsh-skill-studio, which extracts reusable skills from your own sessions: the recommender looks outward at the open-source ecosystem, while the studio looks inward at what you already do well.