← ClaudeAtlas

framework-radarlisted

Research coding agents, agent SDKs, orchestration frameworks, harnesses, protocols, evaluations and relevant videos, then execute every selected hypothesis against the current repository. Use when asked what is new, what agent practices are worth testing, or to run the framework radar.
djnsty23/claude-auto-dev · ★ 3 · AI & Automation · score 65
Install: claude install-skill djnsty23/claude-auto-dev
# Framework Radar Collect first, judge second, test third. The collector owns source retrieval, transcript storage, population counts and deduplication. This skill owns relevance, corroboration, controlled experiments and evidence-backed adoption. ## 1. Collect Run the shipped collector from this plugin: ```bash node "${CLAUDE_PLUGIN_ROOT}/scripts/framework-radar.js" --days 14 ``` If the invocation includes a YouTube URL or ID, append one `--video <ID-or-URL>` for each. A supplied video is always considered even when it falls outside the discovery window. The collector prints: - The number of configured, successful and failed sources. - The primary, research, community, category, video, transcript and retained-comment populations. - `COULD NOT CHECK` for each unavailable source. - The absolute path to its JSON manifest. - The absolute path to a self-contained HTML findings dashboard, with adjacent Markdown and JSON versions under `.claude/reports/`. It uses `YOUTUBE_API_KEY` when available. Otherwise it tries `yt-dlp`, then `uvx --from yt-dlp`. Transcript extraction similarly uses `youtube-transcript-api` directly or through `uvx`. Comment sampling uses the YouTube Data API or a bounded yt-dlp fallback. Missing tooling is a named source failure, never an empty success. Completion: a manifest exists and its population line matches the JSON counts. If every source failed, write a short failure report and stop without marking anything reviewed. ## 2. Read the evidenc