Agents365-ai

Production-grade Claude Code skills — diagramming, visualization, and more

Draw.io diagrams — generate .drawio XML, export to PNG/SVG/PDF via local CLI, with visual review loop and editable embedded XML

Mermaid diagrams — text-based charts with auto-layout, GitHub-native rendering, and CLI export to PNG/SVG/PDF

Excalidraw hand-drawn style diagrams — whiteboard aesthetics, exported to PNG/SVG via CLI

PlantUML diagrams — UML, sequence, class, ER, and C4 architecture diagrams via text-based DSL

Tldraw whiteboard-style diagrams — infinite canvas, hand-drawn feel, exported to PNG/SVG

Semantic Scholar paper search — keyword/boolean/snippet search, citation traversal, similar-paper recommendations, author lookup, BibTeX/Markdown/JSON export, with built-in rate limiting and single-script execution

Download scientific paper PDFs by DOI (or title) — 7-source fallback chain (Unpaywall, Semantic Scholar, arXiv, PubMed Central, bioRxiv/medRxiv, publisher direct, Sci-Hub), batch mode, idempotent retries, structured JSON envelope, and SSRF-safe downloads

End-to-end academic research pipeline — 8-phase (Phase 0..7) script-driven workflow across 7 federated sources (OpenAlex, arXiv, Crossref, PubMed, DBLP, bioRxiv, Exa), cross-source dedup, transparent ranking, dual-backend citation chasing (OpenAlex + Semantic Scholar), parallel deep-read fan-out, mandatory self-critique, and cited reports across 5 archetypes

Ai2 Asta MCP — Semantic Scholar academic graph over MCP (no Python). Intent-to-tool routing, safe fields defaults to avoid context blowups, citation traversal, author lookup, snippet evidence retrieval, and DOI/arXiv/PMID via externalIds.

Journal name abbreviation lookup — ISO 4 + MEDLINE, multi-source cascade (JabRef ~25K journals → AbbrevISO → NLM Catalog), BibTeX field rewrite, batch and NDJSON streaming, agent-native JSON envelope with stable error codes, dry-run on every mutating command, --idempotency-key for replay-safe bib rewrites, and atomic cache rebuild that preserves the old cache on failure

Multi-source drug-target due-diligence — turn a ranked gene list (e.g. scRNA-seq DE output) into a per-gene dossier across UniProt, OpenTargets, and PubMed, plus a local cross-lineage DE scan, then re-rank by a configurable composite score (cross-lineage convergence + druggability + disease genetics + tractability + novelty). Disease-agnostic: configurable focus disease and cell-context queries.

Drive the Pi coding agent from Claude Code — model-agnostic, works with whatever provider pi is configured for (DeepSeek, OpenAI, Anthropic, Google, Ollama, LM Studio, or any OpenAI-compatible endpoint). /pi:review with structured findings, /pi:adversarial-review for design challenges, /pi:rescue with background job control (status/result/cancel), and an optional stop-time review gate. No OAuth (API key only). 1:1 fork of openai/codex-plugin-cc with the runtime swapped from Codex to Pi.

Automated topic-driven video podcast creation — research → script → TTS (7 backends: Edge / Azure / Doubao / CosyVoice / ElevenLabs / Google / OpenAI) → 4K Remotion render → BGM mix → Remotion-native subtitles. Multi-platform: Bilibili / YouTube / Xiaohongshu / Douyin / WeChat Channels. Outputs horizontal long-form (16:9 3840x2160) and vertical shorts (9:16 2160x3840). 15-step workflow with mandatory Studio preview, agent-native JSON envelope CLI, and per-video isolation via --public-dir.

Bilibili anime frame & character organizer — download a bangumi/UP video (or local file), extract scene-change keyframes, split scenery vs character crops (anime person detection), then either cluster everyone by CCIP identity or pull ONE character out via a reference folder (one-vs-rest, distance-prefixed matches). Agent-native JSON envelope CLI with --dry-run, --schema, stable exit codes; optional OCR+LaMa subtitle/watermark removal.