Using NineA
Use the filesystem for discovery and commands for execution. Treat every 9A-managed Skill as read-only; change its YAML or upstream provider instead of editing projected files.
Choose the workflow
- Find or run a capability: use
9a search, inspect the projected schema, then pipe one JSON object to itsinvokeentry. - Add one or more JSON APIs: read
references/declarative.md, author YAML, validate it, then use9a add. - Connect MCP, A2A, or another protocol: read
references/integrations.md. - Repair, update, inspect, or remove a workspace view: read
references/troubleshooting.md. - Upgrade the installed software: read
references/troubleshooting.md, explain the difference betweenbrew upgradeand9a update, and obtain user approval before changing packages or restarting the daemon.
Prefer projected invoke commands over constructing direct provider requests.
Do not place credentials in YAML, prompts, command arguments, or projected
files; provide them to ninead through its environment.
Fast path
9a status --json
9a search "what the user needs"
9a project add <capability-id> .agents/skills
printf '%s\n' '<json-input>' | \
.agents/skills/<projected-skill>/scripts/invoke
Use 9a calls start instead of synchronous invocation when work must outlive
one CLI request. Never grant an agent broader ACLs than the requested task.