This collection turns vague requests into durable, reviewable artifacts. It emphasizes evidence, explicit approval gates, reproducible prompts, deterministic helper scripts, and QA before handoff.

Quick start

Browse the available skills:

npx skills add arkgum/arkgum-agent-skills --list

Install one skill:

npx skills add arkgum/arkgum-agent-skills@arkgum-research-to-page

See all installation options, including global, agent-specific, and manual installation.

Skills

Skill What it does Best for Requirements
arkgum-research-to-page Runs source-grounded NotebookLM research, ranks page opportunities, and produces a cited brief plus builder-ready prompt. Landing pages, research pages, product narratives, content-gap exploration. NotebookLM MCP or compatible NotebookLM CLI.

Example prompts

Use $arkgum-research-to-page to research how an Obsidian vault can become
a personal LLM Wiki, then produce a builder-ready landing-page specification.

More examples are available in examples/prompts.md.

Why this collection

  • Grounded before generated. Claims are traced to sources, and uncertainty stays visible.
  • Artifacts over chat residue. Research reports, manifests, prompts, briefs, and QA files survive the session.
  • Human gates where judgment matters. Visual plans and irreversible actions require approval.
  • Deterministic where repetition matters. Scripts handle scaffolding, validation, normalization, and packaging.
  • Portable skill structure. Every skill follows the open Agent Skills specification.

Repository structure

arkgum-agent-skills/
├── skills/                 # Installable, self-contained Agent Skills
├── scripts/                # Repository-level static validation
├── examples/               # Copy-ready prompts
├── docs/                   # Installation, quality bar, and launch notes
├── .github/                # CI and community templates
├── catalog.json            # Machine-readable skill catalog
└── README.md

Each installable skill contains a required SKILL.md and only the resources needed at runtime:

skill-name/
├── SKILL.md
├── agents/                 # Optional UI metadata
├── scripts/ or tools/      # Deterministic helpers
├── references/             # Documentation loaded on demand
└── assets/                 # Templates and reusable output resources

Quality standard

Every published skill must be specific, inspectable, and safe to install. CI checks naming, frontmatter, broken local links, accidental personal paths, and common secret patterns. See docs/quality-standard.md.

Skills can execute tools and scripts. Review a skill before installing it, understand its external dependencies, and grant only the permissions needed for the current task.

Contributing

Bug reports, documentation improvements, compatibility fixes, and focused skills are welcome. Read CONTRIBUTING.md before opening a pull request.

License

Released under the MIT License. Third-party services and companion skills keep their own terms and licenses.

If this workflow saves you research or production time, consider starring the repository. It helps other people discover the project.