SkillMM · Agent Skills Radar

Agent Skills, Claude Code Skills, and Codex workflows are growing across many repositories, but their documentation and runnability vary widely. Stars provide popularity context; they do not explain triggers, dependencies, boundaries, or reusable scenarios.

SkillMM therefore defines success as reducing discovery and judgment cost, not maximizing collection size. Within minutes, users should understand what a Skill does, why it is worth attention, and what to inspect before using it.

Each candidate is deduplicated, verified to contain a real SKILL.md, and reviewed through repository structure, documentation, and key files. Assessment looks beyond popularity to completeness, usability, utility, and maintenance.

Rating, curation score, recommendation rationale, tags, use cases, and pre-install checks form the judgment interface. Scores organize attention without pretending that editorial assessment is a universal objective truth.

Behind the website is a content supply chain: Codex discovers candidates daily, assists deduplication and assessment, writes local Obsidian records, updates daily and overview indexes, rebuilds the data, and publishes the public view.

The three views serve different jobs: Daily Picks answers what matters today, Quality Archive preserves durable value, and All Skills supports systematic retrieval. Freshness, synchronization, and operating reliability matter more than one-off interface novelty.

Local Markdown records preserve assessment detail, sources, and long-term indexes as the single source of truth. The public site receives only rebuilt static JSON and pages, hides local Obsidian paths, and exposes no writes, CMS, or cloud crawler.

This boundary keeps changeable, judgment-heavy research local while Cloudflare Worker serves stable, sanitized, fast public results. Visitors need no account and cannot reach local notes or temporary files.

I led product positioning and knowledge operations: defining curation boundaries, assessment dimensions, taxonomy, information architecture, search and sorting logic, and the daily workflow from local records to public publishing.

Codex assists project discovery, repository reading, deduplication, field structuring, assessment, and site builds. I remain responsible for selection criteria, final ratings, public boundaries, and online acceptance.