trading-analysis

Makes one daily hedging decision for a fixed ordered stock pair, or selects the fixed pair on the run's first day, by querying offline data through the `hedging_mcp` server. Each invocation handles exactly one target date and upserts one recommendation into `results/hedging/`. The same skill can be driven by an external date-loop for any contiguous run window (benchmark, live, or otherwise). Use this skill whenever the user asks for a hedging decision, pair selection, or one daily hedging step.

Generates one standalone WEEKLY equity research report for one symbol on one target week-ending date. Uses the report_generation_mcp server to read offline DuckDB market, news, filings, and peer benchmark data, then writes one Markdown report plus a run-level summary JSON.

Audits XBRL numeric facts in SEC-style filings by comparing the reported (book) value from the instance document against the correct (true) expected value derived from the filing's calculation linkbase, US-GAAP taxonomy, and XBRL sign conventions. The correct value may be a summation recomputation, a sign correction for directional concepts (expenditures, losses must be positive), or an algebraic derivation. Handles fact extraction, context resolution, period matching, balance-type checking, and writing the final JSON result to results/auditing/. Use this skill whenever the user asks to audit a filing, verify a reported XBRL value, compute a calculated value from linkbases, or check numeric consistency in a 10-K or 10-Q — even if they phrase it as "what is the reported value of X", "audit this concept", "check the filing math", or "verify AssetsCurrent for FY2021".

Makes a single daily BUY/SELL/HOLD decision for one stock on a given target_date by querying an offline DuckDB via MCP tools. The skill is driven externally: each invocation handles exactly one (symbol, target_date) pair and upserts the result into an action-list JSON file. Data access is via the `trading_mcp` server, which reads from an offline DuckDB. The same skill powers both backtest-style replay (caller loops over historical dates) and live trading (caller passes today's date, or omits it so the skill uses the latest date available in DuckDB). Use this skill whenever the user asks you to make a trading decision for a single stock on a specific date, run a live trading step, or append one record to a trading action list — phrased as "trade AAPL on 2025-03-05", "decide TSLA today", "run trading for MSFT 2025-04-10", or just "trade NVDA".