The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes several internal Python scripts (e.g., emerging_scan.py, weekly_scout.py, auto_research.py) to perform quantitative market analysis and manage the skill's state. These scripts are invoked via shell commands within the agent's environment.
[EXTERNAL_DOWNLOADS]: The skill connects to external financial data providers and news organizations. It uses the yfinance library to fetch market data and browser tools (like chrome-use or playwright) to access Seeking Alpha, Wall Street Journal, Financial Times, and SEC EDGAR for research purposes.
[DYNAMIC_EXECUTION]: The auto_research.py script implements a self-modification loop where the agent can programmatically edit its own SKILL.md file. This process is governed by a local scoring rubric (evals/RUBRIC.md) and a 'keep-if-improved' logic, allowing the skill to optimize its own instructions based on performance on test cases.
[INDIRECT_PROMPT_INJECTION]: Because the skill is designed to ingest and process large volumes of data from external websites (financial journalism), it possesses an inherent attack surface for indirect prompt injection. The skill implements a 'Skeptic Filter' and strict factual extraction rules as defensive measures to ensure reasoning is grounded in extracted data rather than embedded narrative instructions.
[PERSISTENCE]: The skill maintains a local SQLite database (~/.local/share/trend-research/articles.db) to store ingested articles and track 'theses' over time. This allows the agent to detect theme convergence across multiple weeks and sources.

trend-stock-research