Multi-Agent Mission Development

A lead architect coordinates several AI agents — each playing to its strength — to deliver one complex mission through Spec Kitty. This walkthrough shows the hand-offs from specification to merge and where the human reviewer stays in the loop.

At a glance: one mission, four contributors (Claude, Gemini, Cursor, and a human reviewer), eight work packages, and a two-week window — driven by the standard /spec-kitty.* command sequence.

Context

Dimension Value
Mission 001-cross-platform-chat-upgrade
Agents Claude (spec/plan), Gemini (data modeling), Cursor (implementation), human reviewer
Goal Ship a cross-platform chat upgrade (web + mobile) with improved reliability in two weeks

Playbook

  1. Specify the mission. The lead runs /spec-kitty.specify with the stakeholder brief. Discovery gates confirm scope, users, and success metrics.

  2. Plan and research. Claude executes /spec-kitty.plan to capture architecture; Gemini runs /spec-kitty.research to gather literature benchmarks.

  3. Generate work packages. /spec-kitty.tasks produces eight prompts across API, UI, and infrastructure. [P] flags highlight parallel-safe work such as documentation updates and telemetry instrumentation.

  4. Assign agents.

    • Claude handles plan updates and reviews.
    • Gemini owns data-model.md updates and research prompts.
    • Cursor implements the chat service changes.
    • The human reviewer tracks tasks/for_review/.

  5. Run the orchestration loop.

    • Manual mode: spec-kitty agent action implement WP## per assigned work package.
    • Automated mode: spec-kitty-orchestrator orchestrate --mission 001-cross-platform-chat-upgrade.

  6. Review completed work. The human reviewer processes for_review prompts via /spec-kitty.review, giving feedback or approving work to move to done.

  7. Accept and merge. Once every work package is in done/:

    /spec-kitty.accept   # Validates mission readiness, records metadata
/spec-kitty.merge    # Consolidates lane branches into local main and cleans up worktrees

Outcome

  • Web and mobile chat surfaces upgraded with consistent reliability guarantees.
  • Zero merge conflicts — agents respected prompt-file boundaries.
  • Dashboard snapshot exported for the sprint report.
Edit this page