AGENTS.md

AGENTS.md - FoxFang 🦊 Agent Specification

This file defines the agent system for FoxFang — your personal AI marketing assistant.

1) Agent Architecture

FoxFang uses a coordinator + specialist pattern to help with marketing tasks:

• OrchestratorAgent (system)

  • Receives requests from chat/CLI
  • Loads your context (preferences, past work, current project)
  • Delegates to specialists and returns final output

• Strategy Lead

  • Campaign planning and brief writing
  • Research synthesis and positioning

• Content Specialist

  • Draft generation and variant production
  • Enforces your tone and style preferences

• Growth Analyst

  • Quality review and optimization suggestions
  • Channel-specific recommendations

2) Memory System

FoxFang maintains three types of memory:

• UserPreferences: Your style, tone, favorite formats
• WorkingMemory: Context for the current session/task
• LongTermMemory: Past work and feedback patterns

All data is stored locally on your machine.

3) Learning Loop

1. You review and provide feedback on outputs
2. FoxFang extracts improvement signals
3. Memory updates with confidence weighting
4. Future outputs incorporate learned patterns

4) Tone Enforcement

Your ToneProfile includes:

• Do/Don't guidance
• Preferred vocabulary
• Style rhythm constraints
• CTA patterns
• Forbidden words

5) Events

• session.started
• message.received
• content.generated
• feedback.submitted
• memory.updated

6) Agent Communication

Agents can delegate work:

MESSAGE_AGENT: Content Specialist | Draft a LinkedIn post about: ...

The orchestrator routes to the appropriate specialist.

7) CLI Command Reference

Before suggesting or running CLI commands, agents should consult docs/commands.md.

• Canonical setup form: pnpm foxfang wizard setup [all|providers|channels]
• Channel-only setup can also use: pnpm foxfang channels setup