The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill aggregates financial news headlines and RSS feed content which are then processed by an LLM to generate market briefings. This constitutes a potential attack surface for indirect prompt injection if an attacker were to place malicious instructions within a news headline.
Ingestion points: RSS feeds in fetch_news.py and API data from providers like Finnhub in earnings.py.
Boundary markers: The prompt construction in summarize.py utilizes Markdown headers to demarcate data sections.
Capability inventory: The skill possesses command execution capabilities via subprocess.run (to call openclaw messaging tools) and network access via urllib.
Sanitization: summarize.py implements a HARDENED_SYSTEM_PROMPT that explicitly instructs the LLM to treat inputs as untrusted and ignore any embedded commands.
[DATA_EXFILTRATION]: The skill performs network requests to various external financial news providers and utility services that are not included in the internal whitelist.
Evidence: summarize.py sends URLs to is.gd for shortening. fetch_news.py and earnings.py make network requests to multiple domains for news aggregation and financial data (e.g., finnhub.io, reuters.com, wsj.com).

finance-news