Anthropic / Claude ecosystem and developer skills

Anthropic and Claude-related tooling and studies surfaced new developer resources and internal data. - 'Awesome Claude Skills' GitHub lists extensible Claude skills for documents, code, and research. - Skills include Word edit with track changes, PDF table extraction, Excel bulk processing. - Anthropic published a study: 132 engineers surveyed, 53 qualitative interviews, internal Claude Code usage analyzed. - Engineers reported work changes tied to Claude Code; some describe 'working to make themselves obsolete.' - Hugging Face introduced hf skills plugin supporting SFT/DPO/GRPO and GGUF export for local deploy via Claude Code workflows. - Anthropic piloted 'Anthropic Interviewer' for capturing public AI opinions. - Developer preferences split: Cursor frequently used for control; Claude Code favored for non-programmers and fine-tuning.
Key insights
Community-curated Claude skills expand Claude's task set
The 'Awesome Claude Skills' GitHub project aggregates a curated index of Claude skill modules that extend Claude’s capabilities across document processing, code development, scientific workflows, and multimedia. The index links to multiple skill repositories, enabling on-demand acquisition and integration of skills such as Word revision edits, PDF table extraction, Excel automation, YouTube subtitle summarization, and local file archival.
Anthropic internal study tracks Claude Code adoption and workforce effects
Anthropic conducted a study involving 132 engineers, 53 in-depth interviews, and internal usage analysis of Claude Code. Findings reported by participants include significant workflow changes and comments that engineers feel as if their work is enabling their own redundancy. The study focused on how Claude Code is used within the company and how it alters engineering and research practices.
Hugging Face plugin enables Claude Code-based fine-tuning and local deployment
A Hugging Face 'hf skills' plugin was noted to support fine-tuning methods including SFT, DPO, and GRPO, automate hardware selection on HF, validate before production, and export models to GGUF for local deployment—integrating with Claude Code workflows for model refinement and deployment.
Tooling preferences: Cursor for control, Claude Code for non-programmers
Multiple posts report that advanced developers prefer Cursor to maintain control over code changes, while Claude Code is seen as more suitable for non-programmers. This reflects varied developer workflows: some prioritize manual oversight and granular control, others prioritize abstraction and higher-level product tasks.