Skills
A skill is a knowledge pack for an agent: a policy, a method, a template. How to create one, attach it to an agent, and who may change it.
A skill is a knowledge pack an agent reads: one instruction file plus any supporting material — reference notes, document templates, checklists. The point is always the same: to turn "how we do things here" from something in a person's head into something the agent uses on every task.
Skills live in the workspace and are reused: one skill attaches to any number of agents, and one agent can carry several skills.
The format is Anthropic Agent Skills: a SKILL.md and its files. So a skill written for Bounc is
understood by other tools, and the other way round.
What a skill holds
- A name — unique in the workspace.
- A description — one sentence about when to apply the skill. The agent reads it when deciding whether the skill is relevant to the task at hand, which makes it matter more than it looks.
- The content — the instructions themselves.
- Files — whatever the skill carries alongside: sample documents, mapping tables, examples.
How to add one
On the Agents page, under the Skills tab, New skill offers three ways:
- Write it by hand — an empty file and your instructions.
- Import by URL — pull in a published skill; ClawHub is supported today, other sources are planned.
- Copy from a runtime — bring a skill already installed on a local runtime into the workspace. Requires a local runtime that is online.
The platform also ships a built-in catalogue of skills about Bounc itself — working with issues, how mentions behave, creating agents, finding and building workflows. They are installed from the CLI and behave like ordinary workspace skills afterwards.
Attaching to an agent
A skill takes effect only once it is attached to an agent: on the agent's page, in the properties on the right, the Skills block and its Add button. The skill's page shows on the right which agents use it.
- Changes are picked up on the next run. A task already under way finishes with the version it started on.
- Connectors bring their own skills. When an agent is given a connector, the description of how to use that access comes with it — nothing to attach separately.
Who may change one
A skill's creator, and the workspace's admins and owners, can edit and delete it; everyone else sees it read-only. If somebody else is editing at the same time, the interface says so and lets you choose: reload their changes, or overwrite with yours.
The skill list has All / Used / Unused / Created by me filters and an agent count per skill — so you can see which skills are genuinely in service and which are worth revisiting or removing.
Skills and connectors
They are easy to confuse, though they complement each other:
- A skill is knowledge. The agent reads it to know how to approach the work: which policy, what tone, what shape the answer takes.
- A connector is access. It lets the agent do something outside: post to Slack, call an API, open a repository.
A good agent usually has both: access to a system, and the rules for using it.
An imported skill can carry executable files. Bounc hands a skill's files to the agent as they are, so read the content and the files before importing, and take skills from sources you trust — especially in workspaces handling sensitive data.
Next
- Agents — what a skill attaches to.
- Connectors — the other half of an agent's equipment.
- Teams — putting specialized agents into one group.