Fabien POLLY

You can manage GitHub repositories, pull requests, issues and code reviews; run GGUF models locally via llama.cpp; geocode addresses and compute routes; extract text from PDFs with OCR; transcribe YouTube videos; track RSS feeds; control Philips Hue lights, browsers, and the desktop; and produce cited research reports from web and arXiv sources.

Software engineers, ML researchers, and power users of the LaRuche environment. The manifest emphasizes test-driven development, systematic debugging, code simplification, spike experiments, and codebase inspection, alongside research-oriented skills like arXiv search, deep research synthesis, and Obsidian note management.

Most skills are self-contained, but some declare explicit command prerequisites: blogwatcher requires blogwatcher-cli, openhue requires the openhue command, and codebase-inspection requires pygount. Computer-control and browser-control need the respective computer and browser runtimes, while app-control uses LaRuche's declared app actions.

The plan skill writes actionable markdown implementation plans without executing them, while long-running-work tracks multi-step efforts using todos, plan mode, kanban, and missions. The cron_manager schedules prompts to run later once or on repeating cron expressions, and delegation hands sub-tasks to specialist models or background jobs.

Yes. llama-cpp runs GGUF models entirely locally, local-git inspects and commits work in isolated worktrees, cognitive-memory stores lasting facts in a local memory map, and Obsidian skills read and edit local vault notes. GitHub skills authenticate via token, SSH key, or the gh CLI under your own credentials.