The Agent Skills Directory

[COMMAND_EXECUTION]: The script collectors/browser_scroll_cdp.mjs connects to a local browser instance using the Chrome DevTools Protocol (CDP). It executes JavaScript commands directly within the browser context via Runtime.evaluate to automate scrolling, button clicking (e.g., 'Show original'), and data extraction from X/Twitter.
[DYNAMIC_EXECUTION]: The collectors/browser_scroll_cdp.mjs script dynamically constructs and executes JavaScript functions to interact with the DOM of the target website. This technique is used to identify tweet structures, extract timestamps, and capture text content during the scraping process.
[EXTERNAL_DOWNLOADS]: The skill is configured to fetch data from several external sources to build its research models, including:
Substack RSS feeds for long-form research articles.
The official X/Twitter API for gathering user timelines.
Apify's managed scraping infrastructure for historical data collection.
Yahoo Finance via the yfinance library to retrieve asset pricing data for signal evaluation.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and analyzes untrusted data from X/Twitter posts to derive investment thesis templates. This creates an attack surface where malicious instructions embedded in a trader's post could attempt to manipulate the agent's analysis or output.
Ingestion points: Data enters the system through scripts/x_trader_builder.py and collectors/stable_collectors.py via various import commands.
Boundary markers: The system attempts to separate a trader's original text from quoted text and applies semantic labels (e.g., keep, deweight, delete) to classify signals.
Capability inventory: The skill can write files (CSV/Markdown), perform network operations (fetching price data), and execute local scripts via sub-commands.
Sanitization: Basic HTML stripping and regex-based extraction are used, but the skill lacks deep semantic sanitization for instruction-like patterns in the raw post text.

x-trader-builder