Outline Skills

An Agent Skills package for operating Outline knowledge bases through outline-cli.

Manage documents, collections, search, comments, attachments, users/groups, sharing, and related workflows. Publish local Markdown with uploaded assets, and create rich-text document references using Outline's mention syntax.

One skill directory, shared across compatible agents. No separate skill format or OpenAI-specific manifest is required. The skill supplies instructions; the Python CLI supplies the API client.

Install

1. Install the CLI

pip install outline-kb-cli

The output controls below require CLI 0.2.0 or the current source checkout. Until that version is published, install from a checkout with pip install -e . in the repository root.

2. Install the skill

Use the cross-agent skills installer (requires Node.js/npm):

npx skills add visualdust/outline-skills --skill outline-skills

Choose your agent in the installer, or specify one explicitly:

npx skills add visualdust/outline-skills --skill outline-skills -a codex
npx skills add visualdust/outline-skills --skill outline-skills -a claude-code

Installation is project-local by default; add -g for user-wide installation. The installer supports agents including Claude Code, Codex, Gemini CLI, OpenCode, GitHub Copilot, and Cursor, and handles their installation directories.

For manual installation, copy the entire skills/outline-skills/ directory, including references/, into your agent's skills directory. Examples: .claude/skills/outline-skills/ for Claude Code or .agents/skills/outline-skills/ for Codex. Installation locations vary; consult your agent's documentation.

Claude Code users can alternatively install the same skill through the existing plugin:

/plugin marketplace add visualdust/outline-skills
/plugin install outline-skills

3. Configure Outline

Create an API key

Open your Outline workspace settings, go to API, and create an API key. Copy the key for the configuration step below; do not share it or commit it to version control.

Configure credentials

Set environment variables:

export OUTLINE_API_KEY="ol_api_..."
export OUTLINE_BASE_URL="https://outline.example.com/api"

The base URL should include /api. Alternatively, create ~/.outline-skills/config.json:

{
  "api_key": "ol_api_...",
  "base_url": "https://outline.example.com/api"
}

On Unix, protect the config with chmod 600 ~/.outline-skills/config.json.

Verify the connection

Run:

outline-cli auth info

If successful, the command returns your user and workspace information. For errors, see Troubleshooting.

Now ask your agent to search Outline, create a document, or manage a collection. Compatible agents can discover the skill from its description; explicit invocation syntax depends on the agent.

CLI examples

outline-cli search "deployment guide" --limit 5
outline-cli documents info --id "document-id" --fields data.id,data.title,data.revision
outline-cli documents info --id "document-id" --text-only
outline-cli documents info --id "document-id" --full-text
outline-cli documents create --title "Report" --collection-id "collection-id" --text-file ./report.md
outline-cli documents update --id "document-id" --text-file ./report.md
outline-cli comments create --document-id "document-id" --data-file ./comment.md

Get only what you need:

• Default summaries are compact JSON without repeated help metadata or expanded document authors.
• --fields selects complete values from the original API JSON; dotted paths work across lists, preserving response shape. No unrelated raw fields are returned.
• documents info --text-only returns exact full Markdown; --full-text includes full Markdown in summary JSON. Neither requires a noisy raw response.
• --max-text-chars 0 suppresses previews, including search context. Never edit from a truncated preview.
• --pretty indents JSON for humans; --raw returns the full API response for debugging.

Errors/warnings go to stderr. Put connection flags such as --base-url before the subcommand. See output controls for field paths and mode constraints.

For Markdown with local images/assets, use documents create-from-file instead of --text-file, starting with --dry-run:

outline-cli documents create-from-file --file ./report.md --collection-id "collection-id" --dry-run

Documentation

• Skill instructions
• Command reference
• Output controls
• Setup and authentication
• Local Markdown publishing
• Troubleshooting

Development

uv run --extra dev python -m ruff check outline_cli tests
uv run --extra dev python -m mypy outline_cli
uv run --extra dev python -m pytest -q

License

MIT.