| name | multica-agent-roster-apply |
| description | Apply a Multica agent roster from a strategy note or explicit spec. Use when the user wants to create, rename, retarget runtime/model, or re-skill Multica agents in bulk from a defined roster. This is a mutating setup skill for Multica agent configuration. |
| disable-model-invocation | true |
Multica Agent Roster Apply
Apply a desired Multica agent roster safely and deterministically.
Use when
- The user wants to rebuild Multica agents from a strategy document.
- The user wants to standardize runtime/model choices across agents.
- The user wants to create missing agents, rename existing ones, or update descriptions/instructions/skill assignments in bulk.
Goal
- Bring the live Multica agent roster into the desired state.
- Minimize accidental churn.
- Leave a clear created/updated/unchanged summary.
Do
- Read the desired roster spec first.
- Probe only what is needed:
multica runtime list --output json
multica agent list --output json
multica skill list --output json
- Resolve the target runtime for each agent before mutating anything.
- Diff desired agents against existing agents by name first, then by obvious continuity clues.
- Create missing agents.
- Update existing agents when any of these drift:
- name
- description
- instructions
- runtime
- model
- visibility
- Set skill assignments explicitly with
multica agent skills set when the desired roster includes skills.
- Verify the final roster with
multica agent list --output json and multica agent skills list <agent-id> --output json.
Do not
- Do not archive or delete agents unless the user explicitly asked for cleanup too.
- Do not trigger issue execution.
- Do not assign issues or rerun tasks as part of roster application.
- Do not assume archived or legacy agents are safe to remove without reference checks.
Mutation boundary
Allowed mutations:
multica agent create
multica agent update
multica agent skills set
Not part of this skill by default:
multica agent archive
- issue assignment
- issue rerun
- autopilot trigger
Inputs
- A strategy summary, roster note, or explicit agent table.
- Desired runtime policy.
- Desired model policy.
- Optional desired skill mapping.
Outputs
- Created agents list.
- Updated agents list.
- Unchanged agents list when relevant.
- Final roster snapshot: name, id, runtime, model, status.
Verification checklist
- Every intended agent exists.
- Every intended agent points to the intended runtime.
- Every intended agent has the intended model string.
- Every intended agent has the intended skill set.
- No unexpected execution-triggering commands were run.
Done when
- The live Multica roster matches the desired roster spec.
- Verification output is captured.
Escalate when
- The required runtime is missing or offline.
- A rename cannot be matched safely to an existing agent.
- Live issues depend on an agent that appears to need replacement rather than update.
- The user also wants archival cleanup of old agents. In that case, hand off to
multica-safe-agent-cleanup after roster application.
Tool policy
- Use only the
multica CLI for Multica reads/writes.
- Prefer
--output json.
- Never print secrets or tokens.
Reporting format
Always report:
- Inputs used for the roster decision.
- Commands category: read-only vs mutating.
- Created / updated / unchanged agents.
- Final verification summary.