audit-strategy-spec

What problem does it solve?

When building quantitative trading strategies via AI agents, unvetted, unconfirmed assumptions in strategy_spec.yaml can lead to invalid, non-reproducible backtest results that waste research compute and produce false positive strategy performance claims.

Core Features & Use Cases

• Field Provenance Classification: Traces conversation history to classify all material spec fields as confirmed, default, unconfirmed, or agent-added, ensuring no unapproved assumptions slip through.
• Component Catalog Validation: Audits all strategy components against the registered component catalog to catch non-canonical decompositions, unregistered custom components, or deviations from standard recipes.
• Backtest Gating: Blocks formal backtest execution until all unconfirmed, default, or agent-added fields are explicitly confirmed by the user, preventing wasted compute on invalid strategy specs. Use Case: A quant researcher using an AI agent to build a momentum strategy can use this skill to verify that all indicator parameters, execution assumptions, and component choices are explicitly approved before running a backtest, avoiding false performance results from unvetted assumptions.

Quick Start

Use the audit-strategy-spec skill to review your strategy_spec.yaml and raw conversation history to identify any unconfirmed assumptions or component provenance issues before running a backtest.