From supervibe
Use WHEN facing 2+ independent tasks BEFORE starting them sequentially to determine if parallel subagent dispatch saves time without coordination cost. Triggers: 'parallel dispatch', 'fan-out', 'параллельно агентов', 'разнеси задачи'.
npx claudepluginhub vtrka/supervibe --plugin supervibeThis skill is limited to using the following tools:
WHEN you have 2+ tasks to do AND you're about to start them. BEFORE you go sequential by default.
Mandates invoking relevant skills via tools before any response in coding sessions. Covers access, priorities, and adaptations for Claude Code, Copilot CLI, Gemini CLI.
Share bugs, ideas, or general feedback.
WHEN you have 2+ tasks to do AND you're about to start them. BEFORE you go sequential by default.
NOT for: single tasks, tasks with sequential dependencies, tasks sharing mutable state.
Parallelism is allowed only when it lowers wall-clock time without increasing merge, context, review, or policy risk. The default for uncertain ownership is sequential execution.
Each dispatched task needs a self-contained worker packet: objective, why it matters, read context, write set, acceptance criteria, verification command, stop condition, output contract, and explicit note that the worker must not revert unrelated edits or assume it is alone in the codebase.
After dispatch, continue until every launched task is collected, reviewed, and
classified as success, blocked, rejected, or partial-with-user-acceptance.
Do not stop after the first subagent returns. A failed subagent does not cancel
independent completed outputs, but it must block the final claim until repaired,
quarantined, or explicitly accepted as partial.
Every successful subagent result must come from a real host Agent/Task call and carry a host invocation id. Do not emulate a subagent result in the parent agent; save non-agent notes separately if the host dispatch did not run.
Parallel work is done only when aggregation has checked write-set conflicts, contradictory recommendations, missing evidence, and verification status for each task. The combined answer must cite per-agent outcomes and explain any discarded, re-run, or serialized task.
Follow docs/references/skill-expert-operating-standard.md: start from source of truth, preserve retrieval evidence, apply scope safety, use real producers with runtime receipts for durable delegated outputs, verify before completion claims, and keep confidence below gate when evidence is partial.
Are tasks independent?
├─ Same files modified → SEQUENTIAL (parallel = merge conflicts)
├─ Same in-memory state → SEQUENTIAL
├─ Sequential output dependency (B needs A's result) → SEQUENTIAL
└─ All criteria pass → PARALLEL OK
If parallel OK, count tasks:
├─ 2-5 tasks → parallel via single message with multiple Agent tool calls
├─ 6-15 tasks → parallel batches of 5
└─ 16+ tasks → use supervibe:subagent-driven-development with full plan
supervibe:confidence-scoring
f. For durable outputs, issue workflow receipts with hostInvocation.source
and hostInvocation.invocationId, then run npm run validate:agent-producer-receiptsReturns:
supervibe:subagent-driven-development — preferred for plan-execution at scalesupervibe:executing-plans — alternative when sequential is fine