The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and process untrusted data from external sources such as social media (Twitter/X, Reddit, Discord) and news headlines.
Ingestion points: Untrusted data enters the agent context through the analyze_tweet function (processing tweet['text']), analyze_reddit_post (processing post titles and comments), and analyze_headline (processing news strings) within references/patterns.md.
Boundary markers: The skill implements basic text cleaning in clean_tweet using regex to remove cashtags, URLs, and mentions. However, it lacks robust delimiters or explicit instructions to the agent to disregard potential instructions embedded within the processed text.
Capability inventory: The skill provides logic for sentiment scoring and data aggregation. No high-risk capabilities such as arbitrary shell execution, network exfiltration, or sensitive file system access were identified in the reference scripts.
Sanitization: Basic sanitization is performed via regex cleaning and truncation of text to 512 tokens for model compatibility, which provides some mitigation against long-form injection attempts.
[EXTERNAL_DOWNLOADS]: The implementation patterns in references/patterns.md rely on the transformers library to load the ProsusAI/finbert model.
The code utilizes the transformers.pipeline API, which downloads pre-trained model weights from Hugging Face's official repository. This is a standard practice for the intended financial NLP tasks.

sentiment-analysis-trading