Everything you need to know about core-skills.
core-skills is a set of 9 modular skills (slash commands) for Claude Code that handle operational documentation. They process meeting transcripts, track tasks, prepare meeting briefings, extract knowledge, and generate daily dashboards -- all from your terminal.
You need Claude Code installed. That's the only requirement. core-skills are plain text files (skill definitions) that Claude Code reads -- no additional software, no database, no server.
Yes, core-skills is open source and free to use. You do need a Claude Code subscription (which includes API usage), but the skills themselves are free.
Claude Code supports user-defined skills via ~/.claude/skills/. Each skill is a folder with a SKILL.md file that defines its behavior. When you type a slash command like /ops, Claude Code reads the skill definition and follows its instructions to process your input. The skills run inside your normal Claude Code session -- no separate process or server needed.
All data stays on your machine as plain Markdown (.md) and YAML (.yaml) files in your Obsidian vault or any local directory. Tasks and insights are stored as per-folder _tasks.yaml and _insights.yaml files alongside your meeting content. Nothing is sent to external servers. The skills only read and write local files.
When you use a skill, the transcript is processed by Claude (Anthropic's AI) as part of your Claude Code session. This follows the same privacy model as any Claude Code interaction. The resulting files are saved locally. Anthropic's privacy policy applies to the AI processing itself.
Yes. Share your vault via any sync mechanism (git, Dropbox, iCloud). Tasks have a private: true/false flag for controlling visibility in shared views. Organization configs can be shared as separate skill repos.
Run /ops status to see all available organization configs, the active config for your current directory, team members, and workflow settings. Run /ops help for a usage guide showing how /ops relates to other skills.
Use /ops for any meeting that belongs to a project or organization. It reads your org config for team recognition, terminology, and workflow rules. Use /transcript for personal calls and ad-hoc recordings when you don't need org-specific processing. When in doubt, use /ops -- it falls back gracefully.
Insights are durable pieces of knowledge extracted silently from your meetings: decisions (with rationale), preferences, learnings, opportunities, and patterns. They're written to _insights.yaml files alongside your meeting CHANGELOGs. Over time, this builds a searchable knowledge base. The visualisation app is the primary way to explore them.
No. The skills work with any directory of Markdown files. Obsidian is just a great way to browse and search the resulting documentation. You could use VS Code, any text editor, or even just the terminal.
Create an org config repo with a {org}.yaml file that defines your team structure, language, terminology, and workflow preferences. Symlink it into ~/.claude/skills/ and /ops will pick it up automatically. See the install page for details.
Yes. The org config controls everything: which summary sections to include, which files to update (CHANGELOG, README, task matrix, meetings index), whether to propagate actions, and whether to auto-refresh the dashboard. You can also override at the project level with .claude/ops-config.yaml.
A local Flask web app that reads your vault files in real-time and displays them as interactive dashboards, kanban task boards, activity heatmaps, and insight timelines. It runs locally on localhost:5050 and has no database -- it reads your Markdown and YAML files directly.
No, it's completely optional. core-skills work perfectly without it. The visualisation app is a bonus for people who want charts, dashboards, and a visual overview of their accumulated data.
The insights extraction feature in the visualisation app supports three AI providers:
Ollama -- run locally for full privacy. No data leaves your machine.
Anthropic -- use Claude models via the Anthropic API.
OpenAI -- use GPT models via the OpenAI API.
Configure your preferred provider in the .env file. Ollama is a great choice if you want everything to stay on your machine.