The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches market data, stock lists, and corporate announcements from well-known financial services including Sina Finance (hq.sinajs.cn), Eastmoney (eastmoney.com), and CNInfo (cninfo.com.cn). These requests are performed using the Python standard library in scripts/top_n_pipeline.py and scripts/ann_query.py to support the skill's primary research function.
[COMMAND_EXECUTION]: The instructions and scripts utilize env -u http_proxy and similar flags to strip proxy settings before executing data collection scripts. This is documented as a requirement to ensure direct connectivity to domestic Chinese financial data servers and avoid connectivity issues associated with local development proxies.
[INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection because it ingests and analyzes untrusted data from external websites (announcement titles, news headlines, and media reports).
Ingestion points: Corporate announcement data fetched in scripts/ann_query.py and media sentiment data mentioned in references/sentiment-evidence.md are passed into the agent team for analysis.
Boundary markers: The skill implements a robust evidence grading system (L1-L3) and strict instructions in SKILL.md to "宁可标注不确定，不可给错误答案" (Rather mark as uncertain than give wrong answers), which serves as a defensive boundary for the LLM's reasoning process.
Capability inventory: The skill has the capability to perform arbitrary HTTP GET/POST requests via urllib and write analysis results to local CSV files.
Sanitization: No programmatic sanitization or filtering of the fetched text is performed; the skill relies on multi-agent validation and human-in-the-loop review to identify inconsistencies.

daymade-sector-research