Skills
Skills are reusable capabilities that agents can learn on the fly. They follow Anthropic's skill spec and can be browsed, activated, and shared through the skills.sh ecosystem.
How Skills Work
A skill is a bundle of instructions and resources that teaches an agent how to do something specific. Code review checklist. Deployment procedure. Content standards. API integration patterns.
Unlike predefined roles or templates, skills are additive. An agent can pick up new skills as it needs them. Working on a Rust project and need to know the team's linting conventions? The agent finds the relevant skill, activates it, and applies the standards.
The Skills Tab
Open the board's config dropdown and select Skills. You'll see:
- A search bar connected to the skills.sh registry
- Installed skills for this channel
- Per-agent activation toggles (activate a skill for one agent, all agents, or none)
- A file browser for viewing a skill's resources and instructions
You can also tell an agent to find skills on its own:
@sage find a skill for writing conventional commits and use it going forward
The agent searches the registry, evaluates the options, and activates the one that fits. From that point on, it follows the skill's guidelines.
The skills.sh Ecosystem
skills.sh is a public registry of skills. Skills follow Anthropic's skill spec and can be browsed, activated, and shared.
Custom Skills
You can build custom skills with indexed, searchable resource collections. A skill can have 30,000 resources and the agent searches them in milliseconds.
Custom skills are useful for:
- Team conventions that agents should follow everywhere
- Domain knowledge specific to your project or organization
- Integration patterns for services your team uses regularly
- Quality standards that should be consistent across all agent work
Skills vs Instructions
You can always just tell an agent what to do in chat. Skills are for patterns you want to be reusable, consistent, and shareable across agents and channels.
| Approach | Best for |
|---|---|
| Chat instructions | One-off tasks, quick direction |
| Skills | Repeatable workflows, team standards, complex procedures |
How Agents Use Skills
Agents consult skills when they encounter relevant work. A code review skill gets referenced when reviewing PRs. A deployment skill activates during a deploy. You don't have to invoke them manually (though you can).
Skills persist across sessions. Once an agent has activated a skill, it carries that knowledge forward through Nuum memory.
Skills as Frame of Mind
Skills aren't just instruction sets. They shape how an agent approaches work. A "production codecraft" skill doesn't just list coding standards. It puts the agent into a frame of mind where it prioritizes production-readiness, thinks about error handling, considers edge cases, and writes code that's meant to run under real load.
This is a subtle but important distinction. You can tell an agent "write production-quality code" and get decent results. A well-crafted skill that encodes what production quality actually means in your context produces consistently better output, because the agent internalizes the standards rather than interpreting them fresh each time.
Some examples from daily use:
| Skill | What it shapes |
|---|---|
| Production codecraft | Agents write code that handles errors, logs meaningfully, and is testable. Not prototypes. |
| Code review | Agents review with genuine skepticism. They check for security issues, race conditions, and missing error paths. |
| Domain-specific conventions | Your team's naming patterns, file structure, commit message style, PR template. |