Claude Trading Skills (AGIPro)

The pack contains 67 skill directories, each verified to hold a SKILL.md; the repo documents every skill as also shipping code examples, use cases and a references/ folder. The README groups them into a table of 15 categories covering market data and APIs, Solana infrastructure, on-chain analysis, technical analysis, backtesting and strategy, portfolio and risk, DeFi, statistical methods, ML for trading, execution, visualization, market microstructure, quant finance, prediction markets, and tax/accounting/compliance. Some of the methodology is substantive: walk-forward-validation opens by explaining why standard k-fold cross-validation fails on financial time series — lookahead bias, autocorrelation, regime dependence, label overlap — then covers rolling and expanding windows, purged cross-validation, combinatorial purged cross-validation and overfit-detection metrics. pumpfun-mechanics is analytical rather than promotional, documenting the virtual constant-product bonding curve and its spot-price math. Install as a Claude Code plugin with /plugin marketplace add agiprolabs/claude-trading-skills, then /plugin install trading-skills@agiprolabs-claude-trading-skills.

Positioning is “Crypto/DeFi-first. Extensible to all of quant finance,” and the weighting shows: a large share of the pack is Solana-specific infrastructure (jito-bundles, shredstream, yellowstone-grpc, solana-rpc, solana-tx-building, raptor-dex, pumpfun-mechanics) and seven skills are US-centric tax and compliance tooling, while the README leaves equities, options and futures coverage as an extension exercise. The README carries an explicit disclaimer: “This is an analysis and research toolkit. Nothing produced by these skills constitutes financial advice. Always do your own research.”

The skills are markdown instructions with embedded code snippets, not executable packages with a test-verified API surface; the README’s “production-ready code” and “tested patterns” claims are self-reported. The project is effectively single-author — the owner account holds 11 commits and one outside contributor a single commit — and that account is a personal GitHub user with no name, bio or company set. The trade name AGIPro appears only in LICENSE.md, the citation block and the marketplace manifest, so the organization behind the work is not independently verifiable. The repository has been dormant since 2026-06-24. One internal inconsistency to note: the README’s prose claims 17 categories while its own table enumerates 15; the count of 67 skills is exact and verified.