MarketPilot AI — Cross-Model Trading Research & Agent Skills
MarketPilot AI combines 17 cross-model agent skills with a Python research engine for market scans, news and filings, entry/exit plans, position sizing, risk checks, scenario replay, walk-forward testing, alerts, journaling and portfolio review. Built for ChatGPT, Claude, Grok, Gemini and Codex; transparent, testable and paper-only; traders decide.
MarketPilot AI
From market discovery to risk review—one transparent, human-led workflow across models.
Watch the 10-second overview · Open the GitHub Pages site · Read the architecture
Workflow preview
The 11 core workflows
| Skill | Purpose |
|---|---|
| Scanner | Daily setups, opening-range breakouts and a filtered public symbol universe |
| News | Headline tags, timestamps and optional SEC filing metadata |
| Entry | Closed-bar trigger checks and stale-data guard |
| Exit | ATR-based stop and target planning |
| Time Machine | Historical replay plus seeded OHLC bootstrap scenarios |
| Risk | Exact-order PASS/BLOCK checks and fingerprinting |
| Budget | Whole-share sizing from risk, fees, slippage and position caps |
| Broker | Local paper ledger only—no live broker connection or order routing |
| Alerts | One-shot, manually checked price alerts |
| Analyst | SEC XBRL metrics and filing metadata |
| Disclosures | U.S. House PTR index, document retrieval and cautious text extraction |
Plus six research modules: Market Regime, Strategy Lab (chronological walk-forward), Portfolio Exposure, Data Quality, Trade Journal and Post-Trade Review.
Quick start
Python 3.10+ is required. Runtime dependencies: none (Python standard library only).
python -m venv .venv
# Windows: .venv\Scripts\activate
# macOS/Linux: source .venv/bin/activate
pip install -e .
# Offline tests
python -m compileall -q src
PYTHONPATH=src python -m unittest discover -s tests -v
# Example deterministic calculations
marketpilot budget --equity 10000 --risk-pct 0.5 --entry 100 --stop 98
marketpilot risk --symbol DEMO --quantity 5 --entry 100 --stop 98 --target 104 --equity 10000 --quote-age-seconds 5
For an end-to-end offline demo, use examples/synthetic_ohlcv.csv; it is generated synthetic data, not historical market data. Regenerate it with python examples/generate_synthetic_data.py. To install reusable skills locally, run python scripts/install_skills.py all for Claude Code plus the shared Agent Skills path (Codex, Grok and Gemini CLI), or choose a dedicated claude, codex, grok or gemini target. A skills-only OpenAI plugin manifest is included for hosts that support this package format. Gemini Apps can import individual skills, with current script/network limits. See model compatibility.
What it can—and cannot—replace
MarketPilot AI consolidates common scanner, alert, journal, sizing, risk-check and backtest workflows that are often spread across paid tools. It does not replace exchange-grade real-time feeds, institutional research terminals, a brokerage, tax software, or regulated financial advice. Yahoo chart/RSS endpoints are unofficial and may lag or throttle; SEC and House data are public records with their own limits.
Safety and evidence
- No live trading: the broker module only records locally confirmed simulated tickets.
- No secrets in prompts or Git: API credentials are not required for core features; SEC identity is read from a local
SEC_USER_AGENTenvironment variable. - No fabricated results: the included tests use deterministic synthetic fixtures. No profitability or accuracy claim is made.
- Simulation is not prediction: same-bar stop/target ambiguity resolves conservatively to the stop; bootstrap assumptions and sample limitations are reported.
Read the data-source limits, safety model, and model compatibility guide before use. Contributors can follow the testing guide and contribution guidelines.
License and credit
Original MarketPilot AI materials are licensed CC BY-NC 4.0. This is a non-commercial Creative Commons license, not an OSI-approved open-source software license and is not specifically designed for software; see LICENSE for the terms. The independent project was informed by the workflow concepts in Jake Schincariol's day-trader-skill, which is MIT-licensed. Full attribution is in NOTICE.md.