neocloud-capex-tracker

Read one source filing and extract structured financial metrics with full provenance. Use this skill when the user asks to "extract capex from the MSFT 10-K", "pull the headline numbers from this filing", "read the annual report and get the financial data", or any variation of extracting specific financial metrics from a single company filing. This skill processes ONE filing at a time to keep context manageable. It reads the relevant sections, extracts the requested metrics with verbatim quotes and section references, validates against XBRL where available, and writes the results to the database.

Walk the `data/_sources/` archive, find any newly fetched company filings sitting in `_raw/` subfolders, and create canonical human-friendly copies under year-level directories using the naming convention `[yyyy.mm.dd][TICKER][PERIOD][FORM].<ext>`. Use this skill whenever the user says anything like "organize the sources folder", "rename the filings", "tidy up the annual report archive", "clean up _sources", or after any batch of `fetch-company-report` runs. Also use this skill proactively if the user is exploring `data/_sources/` and notices raw unrenamed files, or if a scheduled nightly sweep is kicking off. This is the only sanctioned writer of canonical filenames — never rename files in `_sources/` by hand or via other tools.

Look up a specific financial metric for a company and period. Returns the value with full provenance (verbatim quote, section reference, source file). Use this skill when the user asks "what was MSFT's FY2025 capex?", "how much did Amazon spend on PP&E?", "show me Google's revenue", or any question about a specific financial data point for a tracked company. The skill checks the extractions cache first and only invokes read-and-extract on a cache miss, so repeated queries are instant.

Fetch the latest authoritative public filing (10-K, 10-Q, 20-F, HKEX annual or interim report) for a tracked company directly from the primary regulator (SEC EDGAR or HKEXnews), save the original bytes into the project's immutable `data/_sources/<TICKER>/_raw/` archive, write a sidecar JSON, and insert a `source_documents` row in the database. Use this skill whenever the user or another agent says anything like "fetch the latest Microsoft 10-K", "grab Nvidia's most recent quarterly", "pull the Tencent annual report", "download the 20-F for Alibaba", or asks to obtain, retrieve, ingest, or load a public-company financial filing from a primary source. Always prefer this skill over ad-hoc web scraping or general web search when the target is a formal SEC or HKEX filing — it is the only sanctioned way into the repo's source-of-truth archive.