Create skills agents can use again and again.
Skills turn repeatable instructions, scripts, examples, and operating rules into capabilities every agent can reuse. This guide walks through the full lifecycle: create the skill, push files, update it safely, select it for agents, then connect it to schedules, triggers, and orchestrations.
Package repeatable work
Turn prompts, scripts, examples, assets, and operating rules into reusable skills agents can discover.
Push updates
Create, update, list, and retire custom skills from the SDK or API as your workflow matures.
Select capabilities
Attach built-in and custom skills to agents so every run starts with the right operating surface.
Automate execution
Schedule recurring work, react to events, and coordinate multi-agent workflows around selected skills.
Design the skill package
A useful skill is more than a prompt. It should explain when to use the capability, what inputs it expects, what output quality means, and which scripts or assets the agent can run.
| SKILL.md | The primary operating manual. Describe trigger conditions, workflow, constraints, examples, and expected deliverables. |
|---|---|
| Scripts | Optional executable helpers in scripts/ for work that should be deterministic, tested, or too tedious for the agent to rewrite. |
| Examples | Reference inputs, expected outputs, templates, or fixtures that make the skill easy to use consistently. |
| Metadata | Name, description, icon, category, and active state so the skill is discoverable in the platform and SDK. |
Create and push the skill
Publish the skill once, then reuse it from agents, projects, scheduled work, and event triggers. Keep skill IDs stable so run configurations can select the same capability over time.
| Create | Use client.skills.create or POST /v1/skills to store the skill, files, category, and active state. |
|---|---|
| Update | Patch the markdown, files, metadata, or active state when the workflow changes. |
| List | List active skills by category when building internal skill pickers or deployment tooling. |
| Retire | Mark a skill inactive or delete it when teams should stop selecting it for new runs. |
Select skills for agents and runs
Skills become operational when they are selected on the agent that will do the work. Built-in skills can be turned on or off, and custom skills can be attached by ID.
| Agent defaults | Set enabledSkills on an agent so every thread, scheduled job, trigger, or orchestration step starts with the same capabilities. |
|---|---|
| Built-in skills | Use built-ins for research, browser work, image generation, slides, frontend design, memory, task management, app platform work, and email. |
| Custom skills | Attach your published skill IDs alongside built-ins for team-specific operating procedures. |
| Platform selection | Users can also select skills from the task input connector menu before starting a thread. |
Automate skill-driven work
Once a skill is selected on an agent, you can use the same agent in schedules, triggers, and orchestrations. This keeps recurring work consistent without repeating long prompts.
| Schedules | Run recurring jobs such as daily research briefs, weekly release reviews, and monthly reporting workflows. |
|---|---|
| Triggers | React to webhooks, GitHub events, Slack, email, cron, or custom events with a selected agent. |
| Orchestrations | Coordinate multiple agents in sequential, parallel, conditional, or map-reduce workflows. |
| Runs | Trigger, test, inspect, enable, disable, and list execution history from the SDK. |
Examples worth building
Skills are most valuable when they remove ambiguity from repeat work. Start with workflows where quality depends on a checklist, house style, tool sequence, or output template.
| Product release review | Inspect release tasks, deployments, logs, secrets, rollback notes, and open blockers before a production launch. |
|---|---|
| Market research brief | Run deep research, capture sources, create charts or images, and write a structured report into the project workspace. |
| Brand content workflow | Apply brand voice, image style, channel rules, and approval criteria whenever an agent creates campaign assets. |
| Data enrichment runbook | Read CSVs or databases, enrich rows with browser research or APIs, validate results, and write structured output. |
| Incident response | Gather logs, summarize impact, draft customer updates, create follow-up tasks, and produce an incident report. |
Manage skills professionally
Treat skills like operational assets. Review them, test them, document ownership, and update dependent agents when behavior changes.
| Ownership | Give each skill a clear owner, category, purpose, and expected output so teams know when to use it. |
|---|---|
| Versioning | Keep change notes in metadata or examples and avoid breaking output contracts without updating dependent automations. |
| Security | Never store API keys in skill files. Use Secrets and resource bindings for credentials and runtime configuration. |
| Testing | Run representative threads, schedule tests, trigger tests, and orchestration dry runs before enabling a skill widely. |