End-to-end marketing automation for Montse Swim, starting with Instagram image posts.
- Design phase: Mostly complete with open gaps — requirements, workflow, agent design, automation phasing, outreach workflow, and brand voice are documented; several decisions remain pending.
- Implementation: In progress — Instagram MCP server stub implemented with synthetic data; orchestration, approval UI, and metrics wiring pending.
- Creative generation pipeline (flat designs → AI lifestyle variants)
- Instagram publishing workflow (via MCP server stub with synthetic data)
- Variant testing with engagement metrics and winner selection
- Influencer outreach (DM-first, email fallback) with manager approval
- Six agents: orchestrator, creative generation, approval, publishing, metrics, media boost
| Directory | Purpose |
|---|---|
.voice-copilot/ |
Voice Copilot design hub — interview artifacts, transcripts, recordings |
.voice-copilot/README.md |
Design phase file index and interview state |
.voice-copilot/artifacts/ |
All design documents from discovery interviews |
.github/agents/ |
Agent definitions (system-builder) |
.github/skills/ |
Reusable skills for building agents, MCP servers, etc. |
.github/prompts/ |
Plan prompts for system architecture |
mcp-servers/instagram-stub/ |
Implemented MCP server stub with fixtures and tool handlers |
- Confirm open interview gaps (brand guideline specifics, engagement scoring weights and thresholds, auto-approve/outreach safety limits, posting cadence windows, influencer selection criteria)
- Build orchestration state machine and event bus
- Create approval UI/workflow
- Wire publishing queue, metrics polling, and winner logic to the stub
- Prepare sample campaign and run end-to-end tests