hrfish

MiroFish

AI Career Simulator powered by Claude Code SubAgents.

Simulate 10-year career trajectories with multi-path analysis, social sentiment simulation, and detailed HTML reports.

What It Does

Input a resume and career context. MiroFish generates a comprehensive career simulation report covering:

• 5 parallel career paths with best/likely/base/worst scenarios per path
• 10-year income projections with interactive charts
• 30 AI agents discussing your career choices from different perspectives
• Fact-checked claims against real market data
• Reskilling recommendations prioritized by cross-path impact

Report Output

Report Sections

Origin

This project is a Claude Code-native fork of 666ghj/MiroFish, the multi-agent social simulation platform.

What we kept from the original:

• Domain models: BaseIdentity, CareerState, life event definitions
• Core engines: LifeEventEngine, BlockerEngine (6-category career blockers)
• Zep knowledge graph integration for candidate memory
• OASIS-inspired social simulation concepts

What changed:

• Replaced OpenAI + Flask backend with Claude Code SubAgent orchestration
• Removed frontend (Vue.js) and Docker -- output is a self-contained HTML report
• Added multi-path expansion (5 paths x 4 scenarios = 20 trajectories)
• Added SNS Agent Swarm (30-50 AI characters discussing career decisions)
• Added Pydantic v2 data contracts with normalizer layer for LLM output variance
• Added fact-checking pipeline (Tavily) and macro trend analysis
• Packaged as standalone Python CLI (cc_layer)

Requirements

• Python 3.11+
• Claude Code (for SubAgent orchestration)

Installation

pip install -e .

# Optional dependencies
pip install -e ".[zep]"      # Zep knowledge graph
pip install -e ".[search]"   # Tavily web search
pip install -e ".[otel]"     # OpenTelemetry tracing
pip install -e ".[all]"      # Everything

Quick Start

Demo Mode (no API keys needed)

Generate a report from bundled sample data:

python -m cc_layer.cli.pipeline_run \
  --session-dir cc_layer/fixtures/samples/session_01 \
  --phase report

Full Pipeline

Check pipeline status and follow guided steps:

python -m cc_layer.cli.pipeline_run \
  --session-dir cc_layer/state/my_session \
  --phase status

The orchestrator guide at cc_layer/prompts/orchestrator.md provides step-by-step instructions for running the complete pipeline with Claude Code.

Architecture

cc_layer/
  app/
    models/          # Domain models (CareerState, BaseIdentity, etc.)
    services/        # Business logic (event engine, blocker engine, etc.)
    utils/           # Shared utilities (logging, retry, validation)
  cli/               # CLI tools (sim_init, sim_tick, path_score, report_html, etc.)
  schemas/           # Data contracts (canonical models, normalizer, validator)
  prompts/           # SubAgent prompt templates
  fixtures/          # Sample session data for testing
  tests/             # Test suite

Pipeline Phases

CLI Reference

All CLIs are self-documenting:

python -m cc_layer.cli.sim_init --help
python -m cc_layer.cli.path_score --help
python -m cc_layer.cli.report_html --help
python -m cc_layer.cli.pipeline_run --help

Testing

python -m pytest cc_layer/tests/ -v

License

MIT