RunAgents Skills¶

RunAgents ships with a public skills library for AI coding assistants that need more than generic project context.

Source folder: skills/ on GitHub.

These skills are external-facing workflow packs. They are written to be reusable across customer environments and to avoid private infrastructure assumptions. Each one captures a specific RunAgents job to be done: deploying catalog agents, onboarding governed tools, designing approval policy, debugging runs, or wiring RunAgents behind interfaces such as WhatsApp, Slack, web apps, and internal portals.

Why skills instead of only templates?¶

The template files in AI Assistant Setup are still useful. They teach an assistant the basic RunAgents commands and project structure.

Skills solve a different problem: they give the assistant a reusable operating workflow for one specific class of RunAgents work.

That means:

• less generic prompting
• more consistent outputs
• safer production workflows
• better handoff across operators and assistants

Design principles¶

The public RunAgents skills library is designed to be:

• assistant-agnostic — usable with Codex, Claude Code, Cursor, and similar tools
• external-facing — written for customer and partner environments, not only internal operators
• workflow-scoped — each skill solves one concrete RunAgents job to be done
• composable — skills can be paired with templates, MCP tools, and action plans
• platform-aware — they assume RunAgents owns identity propagation, policy, approvals, consent, and tool auth

Coverage review¶

The current library is broad across the core RunAgents lifecycle:

• Build — authoring agents and plan-driven changes
• Wire — catalog deployment, tools, identity providers, and model providers
• Govern — approval policy and OAuth consent debugging
• Operate — run debugging and observability triage
• Interface — web, WhatsApp, Slack, internal portals, and other user-facing surfaces
• Connectors — policy, approval, and observability integrations with external systems

That is a strong first-party baseline for public use. The highest-value future additions would be self-hosted rollout, org-wide governance rollout, and incident-response playbooks, but the current set already covers the most common external deployment and operations paths.

Available skills¶

Build¶

Wire¶

Govern¶

Operate¶

Interface¶

Connectors¶

Use with Codex and skill-native environments¶

If your assistant supports local skill folders, clone the repository and copy the skills you want into your local skills directory:

git clone https://github.com/runagents-io/runagents.git
mkdir -p ~/.codex/skills
cp -R runagents/skills/runagents-approval-policy ~/.codex/skills/
cp -R runagents/skills/runagents-surface-integration ~/.codex/skills/

Then invoke them explicitly in your prompt, for example:

Use $runagents-approval-policy to design a safe approval flow for this Google Workspace assistant.

Use with Claude Code¶

Claude Code does not use the same local skill-folder format as Codex, but the public RunAgents skills still map cleanly into Claude Code using project memory and custom slash commands.

Option 1: Import a skill into CLAUDE.md¶

# RunAgents workflows
@skills/runagents-approval-policy/SKILL.md
@skills/runagents-surface-integration/SKILL.md

This works well when you want those workflows available throughout a project.

Option 2: Create project slash commands¶

mkdir -p .claude/commands/runagents
cat > .claude/commands/runagents/approval-policy.md <<'EOF'
Review @skills/runagents-approval-policy/SKILL.md and apply it to this request: $ARGUMENTS
EOF

Then use it inside Claude Code like:

/runagents/approval-policy Design approvals for Google Workspace calendar writes

Use with Cursor and other assistants¶

If your assistant does not support native skill folders, you can still use the same material.

Recommended options:

1. paste the relevant SKILL.md into project context
2. turn the workflow into a project rule file such as CLAUDE.md, .cursorrules, or AGENTS.md
3. pair it with the RunAgents MCP server so the assistant can act on live workspace data

Recommended starting sequence¶

If you want the strongest first production path, use the skills in this order:

1. runagents-catalog-deploy
2. runagents-tool-onboarding
3. runagents-identity-provider-setup
4. runagents-model-provider-setup
5. runagents-approval-policy
6. runagents-surface-integration
7. runagents-run-debugging

That mirrors a real rollout path:

• deploy a production-shaped agent
• wire tools and identity
• govern writes
• connect the user-facing surface
• debug live behavior with evidence

Strong first example: Google Workspace assistant¶

The Google Workspace assistant is the best first example for these skills because it combines:

• delegated-user OAuth
• policy-controlled writes
• approval and consent flows
• resumed execution
• multi-surface usage
• real business actions such as email, calendar, and document work

Related¶

• AI Assistant Setup
• External Assistants
• Agent Catalog
• Approvals
• Run Lifecycle