A skill is a named piece of Markdown — guidance, conventions, or context — that gets injected into an agent's system prompt when it runs. Skills let you teach the model how you want work done once and reuse it everywhere, instead of rewriting the same instructions into every prompt.
Managing skills
Create and edit skills on the Skills page. Each is stored as a plain .md file in
Weave's data directory (~/.config/weave/skills/ on Linux), so they're easy to
back up or keep under version control.
On weaves & tasks
Skills are selected per weave or task, and a project can set defaults that new weaves inherit. A task's "Override skills" toggle replaces the weave's list for that task only — it's all-or-nothing, not additive, the same way tool and MCP server overrides work.
On agents
An agent can also carry its own baseline skills, set on the Agents page. These are always injected into that agent's system prompt no matter what a weave, task, or chat selects or overrides — they represent what the agent always knows how to do, like a judge agent's scoring conventions or a language specialist's style guide. Agent skills are a floor, not a default you can turn off elsewhere: in the weave/task skill picker, a skill the assigned agent already carries shows as locked.
In chat
A chat attached to a project inherits that project's default skills, the same way a new weave does. Open the chat's settings panel to see them, or switch to "Customize" to pick a different set for that conversation only. Whatever the chat's agent contributes stays on regardless, shown locked in the picker for the same reason as on a task.
Wherever skills apply, the final list sent to the model is the agent's own skills plus whatever the weave/task or chat resolves to, with duplicates removed — so the same skill assigned to both a project and its agent is only injected once.
