The Agent Skills Directory

[SAFE]: The skill implements a robust 'Review Protocol' (review-protocol.md) consisting of audit, challenge, and consistency checks. This protocol serves as a defensive mechanism to ensure the agent's output is anchored in quantitative data and resistant to logical errors or external biases.
[INDIRECT_PROMPT_INJECTION]: The skill ingests external content via financial news and report tools (e.g., get_news_and_reports, get_market_news). While reading untrusted external data is a potential injection surface, the skill mitigates this by enforcing strict data-anchoring requirements and a mandatory 'Devil's Advocate' review phase (R3) that questions the validity of external narratives.
[DATA_EXPOSURE]: No hardcoded credentials, sensitive file path access (such as .ssh or .aws directories), or unexpected environment variable exposure were found. All network operations are directed at financial data tools within the expected scope.
[REMOTE_CODE_EXECUTION]: There are no instructions or scripts that attempt to download, compile, or execute remote code from unverified sources. The mention of backtest_quant.py in the documentation refers to local research logs rather than dynamic runtime execution.
[OBFUSCATION]: The skill's content is clear and uses standard Markdown formatting. No hidden URLs, encoded payloads, zero-width characters, or homoglyph-based spoofing techniques were detected.

stock-main