| name | multica-cli |
| description | Operate the Multica CLI for human + agent project management. Use when the user wants to set up Multica, inspect workspaces/runtimes/agents/issues/projects, create or update Multica issues, assign work to agents, rerun/cancel agent tasks, manage skills, or inspect/autopilot Multica workflows. |
| license | MIT |
| metadata | {"source-docs":["https://multica.ai/docs/cli"]} |
Multica CLI
Use the multica CLI as the source of truth. Prefer CLI commands over direct HTTP calls to the Multica API.
Safety model
Classify the requested operation before running commands.
Read-only: run without asking again
multica version, multica config show, multica auth status
multica daemon status, multica daemon logs, multica daemon disk-usage
multica workspace list/get, multica runtime list/usage/activity
multica issue list/get/search/runs/run-messages, multica issue comment list
multica agent list/get/tasks, multica project list/get, multica skill list/get, multica autopilot list/get/runs
Use --output json when available. For large JSON, summarize the relevant fields instead of pasting everything.
Mutating: require explicit user intent
Proceed if the user clearly asked for the mutation. Otherwise ask a short confirmation.
- Setup/login/logout/config changes:
setup, login, auth logout, config set
- Creating/updating/deleting/archive/restore of workspaces, agents, projects, labels, skills, squads, autopilots
- Issue changes: create/update/status/assign/comment/label/metadata/subscriber
- Daemon lifecycle: start/stop/restart/update
Execution-triggering: require clear target + confirmation
These can start or interrupt agent work. Always inspect the target first, summarize, then confirm unless the user explicitly named the exact target and action.
multica issue assign <id> --agent ...
multica issue rerun <id>
multica issue cancel-task <id>
multica autopilot trigger <id>
Never print full PATs or secret environment values. If command output includes a token, mask it.
Initial probe
When starting a Multica task, run only the probes needed for the request:
multica version
multica config show
multica auth status
multica daemon status
If output says No server configured, use:
multica setup cloud
For self-hosted setups, ask for server-url and app-url, then use:
multica setup self-host --server-url <api-url> --app-url <app-url>
If setup opens a browser and waits, tell the user to complete authentication and then poll the running command.
Common inventory commands
multica workspace list --output json
multica runtime list --output json
multica agent list --output json
multica project list --output json
multica issue list --output json --limit 20
multica skill list --output json
multica autopilot list --output json
Summarize counts and names/statuses:
- workspace slug/name
- daemon running/stopped
- runtime provider/name/status
- agent name/status/runtime/provider
- project title/status/issue count
- issue identifier/title/status/assignee/priority
- autopilot title/status/last run
Issue workflows
List and inspect
multica issue list --output json --limit 20
multica issue list --status todo --output json
multica issue search "README" --output json
multica issue get DOT-33 --output json
multica issue comment list DOT-33 --output json
multica issue runs DOT-33 --output json
Issue IDs may be issue keys like DOT-33 or UUIDs.
Create an issue
For short descriptions:
multica issue create --title "Title" --description "Short description" --priority medium --status backlog --output json
For multi-line or Japanese descriptions on Windows, prefer a UTF-8 file:
Set-Content -Path .pi/tmp/multica-issue.md -Value $Body -Encoding UTF8
multica issue create --title "Title" --description-file .pi/tmp/multica-issue.md --project <project-id> --status backlog --output json
After creating, report the issue identifier and title.
Update status or assignment
Inspect first:
multica issue get DOT-33 --output json
Then mutate:
multica issue status DOT-33 --set todo --output json
multica issue assign DOT-33 --agent "Multica Helper" --output json
Assignment may trigger work depending on workspace rules. Treat it as execution-triggering.
Run or monitor agent work
Before rerun, inspect current issue and current assignee:
multica issue get DOT-33 --output json
multica issue runs DOT-33 --output json
Start a fresh task for the current agent assignment:
multica issue rerun DOT-33 --output json
Monitor:
multica issue runs DOT-33 --output json
multica issue run-messages <run-id> --output json
multica agent tasks "Multica Helper" --output json
Cancel only with explicit confirmation:
multica issue cancel-task DOT-33 --output json
Agent workflows
List runtimes before creating an agent:
multica runtime list --output json
Create agent:
multica agent create --name "Agent Name" --runtime-id <runtime-id> --description "..." --instructions "..." --model "..." --visibility workspace --output json
Prefer --custom-env-file or --custom-env-stdin for secrets; never put real secrets in shell history.
Useful commands:
multica agent get <slug-or-name> --output json
multica agent update <slug-or-name> ... --output json
multica agent tasks <slug-or-name> --output json
multica agent skills <command> ...
multica agent archive <slug-or-name> --output json
Project workflows
multica project list --output json
multica project get <project-id> --output json
multica project create --title "Project title" --description "..." --output json
multica project status <project-id> --set planned --output json
multica project resource <command> ...
When creating issues for an existing project, pass --project <project-id>.
Skill workflows
Multica skills are workspace resources, separate from local Pi Agent Skills. Avoid confusing them.
multica skill list --output json
multica skill get <skill-id> --output json
multica skill create --name "skill-name" --description "..." --content "..." --output json
multica skill import <url> --output json
multica skill files <command> ...
Attach/detach skills through multica agent skills ....
Autopilot workflows
multica autopilot list --output json
multica autopilot get <autopilot-id> --output json
multica autopilot runs <autopilot-id> --output json
multica autopilot trigger <autopilot-id> --output json
multica autopilot trigger-add <autopilot-id> ...
Manual trigger can create issues or run agents. Treat it as execution-triggering.
Reporting format
For each operation, report:
- command category: read-only / mutating / execution-triggering
- command(s) run, redacting tokens/secrets
- result summary: identifiers, statuses, counts, next action
- any blocked step, especially browser auth or daemon stopped
Keep user-facing output concise. Include exact issue keys, agent names, runtime names, and project IDs when relevant.