Agent skills
The skill packages that teach an agent how to work with a Synap pod, and how they are installed.
A skill is a folder an agent reads to learn a job: a SKILL.md with a short description that tells the agent when to
use it, plus reference files it reads only when the task needs them. Synap's skills teach an agent how to work with
your pod: what to store where, how to extend the schema, how to build views, and how proposals work.
The skills are open source, in the skills/ folder of the
backend repository. skills/manifest.json is the list of packages and says which are installed by default.
Baseline skills
Installed by default for every agent that reads skill folders.
| Skill | Use it for |
|---|---|
synap | Core data work: capture, remember, find and link things in your pod. "Save this", "Find my notes on X", "Who is Y". |
synap-schema | Extending the data model: a new kind of thing, or new fields on an existing kind. |
synap-ui | Building interface over your data: views, dashboards, bento layouts and whole workspaces. |
synap-design | Designing and reviewing user interface for Synap or for a UI over your pod. |
Workflow skills
Installed only when you or the client asks for them, because each is for one kind of task.
| Skill | Use it for |
|---|---|
onboard | Setting up a workspace or project that has little or no data yet. |
agent-os | Provisioning workspaces from templates, for example "add a CRM". |
synap-app | Building your own app on Synap with the SDK. |
synap-market | Finding, installing, authoring and publishing marketplace templates, capabilities, automations and cells. |
How skills are installed
synap connect <client> and synap init install the baseline skills for clients that read skill folders, such as
Claude Code (~/.claude/skills/) and Codex. The CLI takes the skills from your pod
(/api/hub/skills/system), so they match the pod's version.
Clients that do not read skill folders, such as Cursor, still get the pod's tools over MCP.
What a skill needs
The skills call your pod over HTTP with the agent's key:
Authorization: Bearer <SYNAP_HUB_API_KEY>
X-Workspace-Id: <workspace id> optionalReads need the hub-protocol.read scope and writes need hub-protocol.write. synap connect sets this up.
How a skill loads
An agent always sees each installed skill's name and description, which costs a few dozen tokens per skill. It reads
the skill's SKILL.md only when the task matches the description, and the reference files only when it needs them.
Installing all the baseline skills is therefore cheap.
Contributing a skill
Some packages are split into topic files: when a package has a _skill.yaml (its front matter) and an _order.txt
(the order of its topic files), its SKILL.md is generated from them by skills/build.mjs. In those packages, edit
the topic files, not SKILL.md. See
Contributing.