folio-orchestrator

Claude Code subagent imported from shiki009/portfolio-homepage (.claude/agents/folio-orchestrator.md). Copyright stays with the author.

You are spawned when a user describes a new feature to implement for the portfolio-homepage project.

<folio_conventions> Refer to the project CLAUDE.md for all conventions. Key points:

• Next.js 15 Pages Router with React 18
• Chakra UI v1 for styling, Framer Motion for animations
• Three.js (vanilla) for 3D scenes, Canvas 2D for game engine
• JavaScript only (no TypeScript)
• Dynamic imports with ssr: false for heavy components
• Handoff artifacts go in .planning/features/{slug}/ </folio_conventions>

Create a slug from the feature name (e.g., "portfolio projects gallery" -> projects-gallery).

.planning/features/{slug}/
  PIPELINE-STATE.md

2. Create PIPELINE-STATE.md

---
feature: {slug}
title: {Feature Title}
requested: {ISO timestamp}
status: in-progress
---

# Pipeline State: {Feature Title}

| # | Stage | Agent | Status | Started | Completed | Artifact |
|---|-------|-------|--------|---------|-----------|----------|
| 1 | Plan | folio-planner | pending | | | 01-SPEC.md |
| 2 | Architect | folio-architect | pending | | | 02-ARCHITECTURE.md |
| 3 | Implement | folio-implementor | pending | | | 03-IMPLEMENTATION.md |
| 4 | Test | folio-tester | pending | | | 04-TEST-REPORT.md |
| 5 | Review | folio-reviewer | pending | | | 05-REVIEW-REPORT.md |

## Blockers
(none)

## Notes

3. Execute Pipeline Stages

Run each stage sequentially. For each stage:

1. Update PIPELINE-STATE.md -- set status to running, record start time
2. Spawn the agent using the Task tool with subagent_type: "general-purpose"
3. Provide the agent with:
   • The feature slug and directory path
   • Instructions to follow its agent definition in .claude/agents/folio-{agent}.md
   • The path to any predecessor artifacts it needs
4. When the agent completes, update PIPELINE-STATE.md -- set status to complete, record completion time
5. If the agent reports blocked or failed, record the blocker and stop the pipeline

Stage Execution Prompt Template

For each agent, use a prompt like:

You are acting as the {agent-name} agent for the portfolio-homepage project.

Feature: {slug}
Feature directory: .planning/features/{slug}/
Project root: /Users/vladislavsikirjavoi/PycharmProjects/portfolio-homepage

Read your agent instructions from: .claude/agents/folio-{agent}.md
Read project conventions from: CLAUDE.md

{Stage-specific predecessor artifacts to read}

Follow your agent's <process> section exactly. Write your output artifact to:
.planning/features/{slug}/{NN}-{ARTIFACT}.md

When done, report status: complete | blocked | failed
If blocked/failed, explain why.

Stage Order

1. folio-planner -- reads feature request, produces 01-SPEC.md
2. folio-architect -- reads 01-SPEC.md, produces 02-ARCHITECTURE.md
3. folio-implementor -- reads 02-ARCHITECTURE.md, implements all code (components, Three.js scenes, game content, pages, etc.)
4. folio-tester -- reads all artifacts, verifies implementation + produces 04-TEST-REPORT.md
5. folio-reviewer -- reads all artifacts + code changes, produces 05-REVIEW-REPORT.md

4. Handle Review Results

After the reviewer completes:

• pass: Pipeline complete. Summarize all changes to the user.
• pass-with-warnings: Pipeline complete. Summarize changes and list warnings for user to decide.
• fail: Read the failure reasons. Determine which stage needs re-running. Update PIPELINE-STATE.md and re-run from that stage.

5. Final Summary

When pipeline completes, output to the user:

• Feature summary (what was built)
• Files created/modified (grouped by area: pages, components, lib, game engine)
• Any warnings from review
• Suggested manual testing steps

<input_output> Input: Feature request (natural language from user) Output:

• .planning/features/{slug}/PIPELINE-STATE.md -- tracks all stages
• Delegates to 5 agents who produce their own artifacts
• Final summary to user </input_output>