Files
Shayan Rais 304eab6e52 replace Oishii example with time orchestrator built by agent teams
Agent teams now bootstrap a self-contained time orchestration workflow
inside agent-teams/.claude/ (command, agent, skill). The prompt, diagram,
and How to Use section all reference the new time-orchestrator pattern
with cd agent-teams && claude → /time-orchestrator.

Co-Authored-By: Claude <noreply@anthropic.com>
2026-03-12 18:30:02 +05:00

3.5 KiB

Create an agent team to build a time orchestration workflow that displays the current Dubai time as a visual SVG card. The workflow follows the Command → Agent → Skill architecture pattern:

  • A command orchestrates the flow and handles user interaction
  • An agent fetches the live current time for Dubai using a preloaded skill
  • A skill creates a visual SVG time card from the fetched data

Important: All files must be created inside agent-teams/.claude/ — NOT in the repo root's .claude/ directory. This keeps the agent team's output self-contained and runnable via cd agent-teams && claude. Do NOT reference or copy the existing weather workflow — build everything from scratch.

Assign these teammates:

  1. Command Architect — Design and implement the /time-orchestrator command in agent-teams/.claude/commands/time-orchestrator.md. The command should:

    • Invoke the time-agent via the Agent tool (NOT bash) to fetch the current time for Dubai, UAE (Asia/Dubai timezone, UTC+4)
    • Invoke the time-svg-creator skill via the Skill tool to render the SVG card from the fetched time data
    • Use model: haiku in the frontmatter
    • Include critical requirements: sequential flow, correct tool usage (Agent tool for agents, Skill tool for skills), and an output summary Coordinate with the other teammates via the shared task list to agree on the data contract ({time, timezone, formatted}) passed between components.
  2. Agent Engineer — Design and implement the time-agent in agent-teams/.claude/agents/time-agent.md and its preloaded time-fetcher skill in agent-teams/.claude/skills/time-fetcher/SKILL.md. The agent should:

    • Fetch the current time for Dubai (Asia/Dubai, UTC+4) using Bash with TZ='Asia/Dubai' date '+%Y-%m-%d %H:%M:%S %Z'
    • Return the time value, timezone name, and formatted string to the command
    • Use frontmatter: tools (Bash), model: haiku, color: blue, maxTurns: 3
    • Preload the time-fetcher skill via the skills: field The time-fetcher skill (agent-teams/.claude/skills/time-fetcher/SKILL.md) should contain the bash command for Dubai time, the expected output format, and set user-invocable: false since it is agent-only domain knowledge. Post the agreed data contract to the shared task list so the Command Architect and Skill Designer can align on the interface.
  3. Skill Designer — Design and implement the time-svg-creator skill in agent-teams/.claude/skills/time-svg-creator/SKILL.md with supporting files reference.md (SVG template + output template) and examples.md (example input/output pairs). The skill should:

    • Receive a time value, timezone, and formatted string from the calling context
    • Create a self-contained SVG time card for Dubai showing the current time
    • Write the SVG to agent-teams/output/dubai-time.svg
    • Write a markdown summary to agent-teams/output/output.md
    • Use the exact time provided — never re-fetch
    • Keep templates in reference.md (SVG markup with placeholders, markdown output template) and example pairs in examples.md Also create the agent-teams/output/ directory for the output files.

All three teammates should create tasks in the shared task list to coordinate the data contract: the agent returns {time, timezone, formatted}, the command passes it through context, and the skill consumes it. Start all three in parallel since the components are independent — they only need to agree on the data interface, not wait on each other's implementation.