Claude Trading Skills started as a personal project to use AI to improve my own trading process.

Claude Trading Skills is a Claude Skills-based trading workflow toolkit for time-constrained individual investors.

It is designed for investors who use long-term investing, ETFs, and dividend stocks as their core, while using disciplined swing trading as a satellite strategy when market conditions are favorable.

The goal is not to outsource buy/sell decisions to AI. The goal is to structure market review, risk management, trade planning, journaling, and continuous improvement. It is open source because the workflows, checklists, and review habits behind better trading decisions can improve through shared practice.

This is not a signal service or a promise of profitability. It is a toolkit for traders who want to build a better decision process.

The project follows a first for self, open for others stance: it is built first as a practical workflow the author uses, then shared openly for others who face similar constraints.

📖 Documentation site: https://tradermonty.github.io/claude-trading-skills/

Project vision: PROJECT_VISION.md

日本語版READMEはREADME.ja.mdをご覧ください。

This repository is for educational, research, and process-improvement purposes only. It is not financial advice, investment advisory service, tax advice, legal advice, a signal service, or a broker execution platform. Trading and investing involve risk, including loss of principal. Past performance, backtests, screens, reports, and AI-generated analysis do not guarantee future results. All trading decisions, position sizing, tax/regulatory compliance, and broker usage are the user's responsibility.

The project is provided under the MIT License, AS IS, WITHOUT WARRANTY.

This repository is designed for:

• Time-constrained individual investors
• Long-term investors who also want disciplined swing-trading upside
• Dividend and ETF investors who want structured portfolio review
• Traders who want to manage risk before finding trade candidates
• Investors who want to journal and improve their decision process

It is not designed for fully automated trading, signal outsourcing, or short-term scalping.

New users should start with one of these operational workflows. Each link points to a machine-readable manifest under workflows/ that names the exact skills, decision gates, and artifacts in order.

See workflows/README.md for how to read a manifest and run it manually. For a one-page "which workflow fits my situation?" guide, see Find Your Workflow (日本語).

If you do not have FMP / FINVIZ / Alpaca subscriptions, start with these five skills and run them manually:

1. market-breadth-analyzer — public CSV breadth scoring; no API key
2. uptrend-analyzer — public CSV uptrend participation; no API key
3. position-sizer — pure calculation; no I/O
4. trader-memory-core — local YAML journaling
5. signal-postmortem — review framework

This path lets you review market conditions, size trades, journal decisions, and review outcomes without paid data APIs. Note: "no API" does not mean "no external data" — these skills still need public CSVs, chart screenshots, or local files. See each skill's integrations: entry in skills-index.yaml for exact input requirements.

Canonical source: skills-index.yaml is the authoritative index of all skills. If this README, CLAUDE.md, or docs disagree with the index, the index is correct. The same applies to multi-skill workflows — workflows/*.yaml is canonical.

• skills/<skill-name>/ – Source folder for each trading skill. Contains SKILL.md, reference material, and any helper scripts.
• skills-index.yaml – Canonical metadata index for every skill (id, category, integrations, workflows back-references).
• workflows/ – Operational workflow manifests for the Core + Satellite routines (canonical, validator-enforced via --strict-workflows).
• skill-packages/ – Pre-built .skill archives ready to upload to Claude's web app Skills tab.
• docs/ – Documentation site content, generated skill pages, and docs/dev/metadata-and-workflow-schema.md (schema spec).
• scripts/ – Repository-level automation, including the schema validator and one-shot bootstrap helper.
• skillsets/ – Purpose-specific install bundles defining required / recommended / optional skills for major goals (4 core skillsets shipped: market-regime, core-portfolio, swing-opportunity, trade-memory; consumed by the Navigator).

1. Download the .skill file that matches the skill you want from skill-packages/.
2. Open Claude in your browser, go to Settings → Skills, and upload the ZIP (see Anthropic's Skills launch post for feature overview).
3. Enable the skill inside the conversation where you need it.

1. Clone or download this repository.
2. Copy the desired skill folder (e.g., backtest-expert) into your Claude Code Skills directory (open Claude Code → Settings → Skills → Open Skills Folder, per the Claude Code Skills documentation).
3. Restart or reload Claude Code so the new skill is detected.

Tip: .skill packages are built from the source folders with tests and local build artifacts omitted. Edit a source folder if you want to customize a skill, then run python3 scripts/package_skills.py --skill <skill-name> before uploading to the web app.

Want a ready-to-run agent-style workflow? See the companion Hermes Trading Research Agent Work Package.

It packages these skills into a Hermes profile with task-oriented slash-command routines such as /pre-market-routine, /after-close-review, /trade-journal, /weekly-portfolio-review, and /monthly-performance-review.

This is a research, journaling, and risk-review assistant, not an automated trading system. It does not place orders, provide a signal service, or run hidden scheduled jobs; human decision gates remain central.

This repository contains skills across the following areas:

The detailed catalog below is auto-generated from skills-index.yaml by scripts/generate_catalog_from_index.py. To update a skill's description, edit its skills-index.yaml entry and re-run the generator (python3 scripts/generate_catalog_from_index.py). For a more navigable version, use the documentation site.

The main Core + Satellite starting path is described above. The examples below show additional ways to compose skills, including advanced satellite and contributor workflows.

1. Use Economic Calendar Fetcher to check today's high-impact events (FOMC, NFP, CPI releases)
2. Use Earnings Calendar to identify major companies reporting today
3. Use Market News Analyst to review overnight developments and their market impact
4. Use Breadth Chart Analyst to assess overall market health and positioning

1. Use Sector Analyst to fetch CSV data and identify rotation patterns (optionally provide charts)
2. Use Technical Analyst on key indices and positions for trend confirmation
3. Use Market Environment Analysis for comprehensive macro briefing
4. Use US Market Bubble Detector to assess speculative excess and risk levels

1. Use US Stock Analysis for comprehensive fundamental and technical review
2. Use Earnings Calendar to check upcoming earnings dates
3. Use Market News Analyst to review recent company-specific news and sector developments
4. Use Backtest Expert to validate entry/exit strategies before position sizing

1. Use Stanley Druckenmiller Investment for macro theme identification
2. Use Economic Calendar Fetcher to time entries around major data releases
3. Use Breadth Chart Analyst and Technical Analyst for confirmation signals
4. Use US Market Bubble Detector for risk management and profit-taking guidance

1. Use Earnings Trade Analyzer to score recent earnings reactions (gap size, trend, volume, MA position)
2. Use PEAD Screener (Mode B) with analyzer output to find PEAD setups (red candle pullbacks → breakout signals)
3. Use Technical Analyst to confirm weekly chart patterns and support/resistance levels
4. Use Liquidity filters in PEAD Screener to ensure position sizing feasibility
5. Monitor SIGNAL_READY stocks for breakout entries with defined stop-loss (red candle low) and 2R targets

1. Use Value Dividend Screener to identify high-quality dividend stocks with sustainable yields
2. Use Dividend Growth Pullback Screener to find growth-focused dividend stocks at attractive technical entry points
3. Use US Stock Analysis for deep-dive fundamental analysis on top candidates
4. Use Earnings Calendar to track upcoming earnings for portfolio holdings
5. Use Market Environment Analysis to assess macro conditions for dividend strategies
6. Use Backtest Expert to validate dividend capture or growth strategies

1. Use Kanchi Dividend SOP to run Kanchi's 5-step process and create buy plans with invalidation conditions
2. Use Kanchi Dividend Review Monitor on a daily/weekly/quarterly cadence to generate OK/WARN/REVIEW queues
3. Use Kanchi Dividend US Tax Accounting to align holdings with qualified-dividend assumptions and account location
4. Feed REVIEW findings back into Kanchi Dividend SOP before adding to positions

1. Use Options Strategy Advisor to simulate and compare options strategies using Black-Scholes pricing
2. Use Technical Analyst to identify optimal entry timing and support/resistance levels
3. Use Earnings Calendar to plan earnings-based options strategies
4. Use US Stock Analysis to validate fundamental thesis before deploying capital
5. Review Greeks and P/L scenarios to select optimal strategy (covered calls, spreads, straddles, etc.)

1. Use Portfolio Manager to fetch current holdings via Alpaca MCP and analyze portfolio health
2. Review asset allocation, sector diversification, and risk metrics (beta, volatility, concentration)
3. Review position-level flags (HOLD/ADD/TRIM/SELL candidates) based on thesis validation
4. Use Market Environment Analysis and US Market Bubble Detector to assess macro conditions
5. Review a rebalancing plan and decide manually which actions, if any, to take

1. Use Pair Trade Screener to identify cointegrated stock pairs within sectors
2. Analyze mean-reversion metrics (half-life, z-score) and hedge ratios
3. Use Technical Analyst to confirm technical setups for both legs of the pair
4. Monitor entry/exit signals based on z-score thresholds
5. Track spread convergence and manage market-neutral positions

• Data Quality Checker (data-quality-checker)

  • Validates data quality in market analysis documents and blog articles before publication.
  • 5 check categories: price scale inconsistencies (ETF vs futures digit hints), instrument notation consistency, date/weekday mismatches (English + Japanese), allocation total errors (section-limited), and unit mismatches.
  • Advisory mode — flags issues as warnings for human review, exit 0 even with findings.
  • Supports full-width Japanese characters (％, 〜), range notation (50-55%), and year inference for dates without explicit year.
  • No API key required — works offline on local markdown files.

• Skill Designer (skill-designer)

  • Generates Claude CLI prompts for designing new skills from structured idea specifications.
  • Embeds repository conventions (structure guide, quality checklist, SKILL.md template) into the prompt.
  • Lists existing skills to prevent duplication. Used by the skill auto-generation pipeline's daily flow.
  • No API key required.

• Dual-Axis Skill Reviewer (dual-axis-skill-reviewer)

  • Reviews skill quality using a dual-axis method: deterministic auto scoring (structure, workflow, execution safety, artifacts, tests) and optional LLM deep review.
  • 5-category auto axis (0-100): Metadata & Use Case (20), Workflow Coverage (25), Execution Safety & Reproducibility (25), Supporting Artifacts (10), Test Health (20).
  • Detects knowledge_only skills (no scripts, references only) and adjusts scoring expectations to avoid unfair penalties.
  • Optional LLM axis for qualitative review (correctness, risk, missing logic, maintainability) with configurable weight blending.
  • Supports --all flag to review every skill at once, --skip-tests for quick triage, and --project-root for cross-project review.
  • No API key required.

• Skill Idea Miner (skill-idea-miner)

  • Mines Claude Code session logs for skill idea candidates, scores them for novelty/feasibility/trading value, and maintains a prioritized backlog.
  • Used by the weekly skill auto-generation pipeline. Can also be run manually.
  • No API key required.

This section is contributor-oriented. New users can skip it and start with the Core + Satellite path above.

An automated pipeline that continuously reviews and improves skill quality. A daily launchd job picks one skill, scores it with the dual-axis reviewer, and if the score is below 90/100, invokes claude -p to apply improvements and open a PR.

1. Round-robin selection — cycles through all skills (excluding the reviewer itself), persisted in logs/.skill_improvement_state.json.
2. Auto scoring — runs run_dual_axis_review.py to get a deterministic score (0-100).
3. Improvement gate — if auto_review.score < 90, Claude CLI applies fixes to SKILL.md and references.
4. Quality gate — re-scores after improvement (with tests enabled); rolls back if the score didn't improve.
5. PR creation — commits changes to a feature branch and opens a GitHub PR for human review.
6. Daily summary — writes results to reports/skill-improvement-log/YYYY-MM-DD_summary.md.

# Dry-run: score one skill without applying improvements or creating PRs
python3 scripts/run_skill_improvement_loop.py --dry-run

# Review all skills in dry-run mode
python3 scripts/run_skill_improvement_loop.py --dry-run --all

# Full run: score, improve if needed, and open PR
python3 scripts/run_skill_improvement_loop.py

The loop runs daily at 05:00 local time via macOS launchd:

# Install the agent
cp launchd/com.trade-analysis.skill-improvement.plist ~/Library/LaunchAgents/
launchctl load ~/Library/LaunchAgents/com.trade-analysis.skill-improvement.plist

# Verify
launchctl list | grep skill-improvement

# Manual trigger
launchctl start com.trade-analysis.skill-improvement

This section is contributor-oriented. It describes repository maintenance automation, not a required trading workflow.

An automated pipeline that mines session logs for skill ideas (weekly) and designs, reviews, and creates new skills as PRs (daily). Works alongside the Self-Improvement Loop to continuously expand the skill catalog.

1. Weekly mining — scans Claude Code session logs for recurring patterns that could become skills, scores each idea for novelty, feasibility, and trading value.
2. Backlog scoring — ranked ideas are stored in logs/.skill_generation_backlog.yaml with status tracking (pending, in_progress, completed, design_failed, review_failed, pr_failed).
3. Daily selection — picks the highest-scoring pending idea; retries design_failed / pr_failed once (but review_failed is terminal).
4. Design & review — the Skill Designer builds a complete skill (SKILL.md, references, scripts), then the Dual-Axis Reviewer scores it. If the score is too low, the idea is marked review_failed.
5. PR creation — commits the new skill to a feature branch and opens a GitHub PR for human review.

# Weekly: mine ideas from session logs and score them
python3 scripts/run_skill_generation_pipeline.py --mode weekly --dry-run

# Daily: design a skill from the highest-scoring backlog idea
python3 scripts/run_skill_generation_pipeline.py --mode daily --dry-run

# Full daily run (creates branch, designs skill, opens PR)
python3 scripts/run_skill_generation_pipeline.py --mode daily

Two launchd agents handle the weekly and daily schedules:

# Install both agents
cp launchd/com.trade-analysis.skill-generation-weekly.plist ~/Library/LaunchAgents/
cp launchd/com.trade-analysis.skill-generation-daily.plist ~/Library/LaunchAgents/
launchctl load ~/Library/LaunchAgents/com.trade-analysis.skill-generation-weekly.plist
launchctl load ~/Library/LaunchAgents/com.trade-analysis.skill-generation-daily.plist

# Verify
launchctl list | grep skill-generation

# Manual trigger
launchctl start com.trade-analysis.skill-generation-weekly
launchctl start com.trade-analysis.skill-generation-daily

• Update SKILL.md files to tweak trigger descriptions or capability notes; ensure the frontmatter name matches the folder name when zipping.
• Extend reference documents or add scripts inside each skill folder to support new workflows.
• When distributing updates, regenerate the matching .skill file in skill-packages/ so web-app users get the latest version:

  python3 scripts/package_skills.py --skill <skill-name>

Several skills require API keys for data access:

Financial Modeling Prep (FMP) API:

• Free tier: 250 requests/day (sufficient for most use cases)
• Sign up: https://financialmodelingprep.com/developer/docs
• Set environment variable: export FMP_API_KEY=your_key_here
• Or provide key via command-line argument when prompted

FINVIZ Elite API:

• Subscription: $39.50/month or $299.50/year
• Sign up: https://elite.finviz.com/
• Set environment variable: export FINVIZ_API_KEY=your_key_here
• Provides fast pre-screening for dividend screeners

Alpaca Trading API:

• Free paper trading account available
• Sign up: https://alpaca.markets/
• Requires Alpaca MCP Server configuration
• Set environment variables:

  export ALPACA_API_KEY="your_api_key_id"
  export ALPACA_SECRET_KEY="your_secret_key"
  export ALPACA_PAPER="true"  # or "false" for live trading

• Claude Skills launch overview: https://www.anthropic.com/news/skills
• Claude Code Skills how-to: https://docs.claude.com/en/docs/claude-code/skills
• Financial Modeling Prep API: https://financialmodelingprep.com/developer/docs

Questions or suggestions? Open an issue or include guidance alongside the relevant skill folder so future users know how to get the most from these trading assistants.

All skills and reference materials in this repository are provided for educational and research purposes.