The Agent Skills Directory

[PROMPT_INJECTION]: The llm_sentiment function in Section 3.1 demonstrates a potential surface for indirect prompt injection.
Ingestion points: Social media text collected from Twitter, Telegram, Discord, and Reddit is passed directly to an LLM provider.
Boundary markers: The prompt template lacks explicit delimiters or instructions to ignore embedded commands within the input text.
Capability inventory: The provided scripts do not include capabilities for arbitrary command execution, file-system modification, or unauthorized network operations based on the LLM output.
Sanitization: The skill implements basic sanitization by truncating the input text to 1000 characters before interpolation.
[EXTERNAL_DOWNLOADS]: The skill recommends the use of standard, reputable Python packages for its operations, including tweepy, telethon, discord.py, praw, and transformers. No unverified or suspicious external dependencies were identified.
[DATA_EXPOSURE_EXFILTRATION]: The skill correctly implements security best practices by using os.getenv to retrieve sensitive API credentials (tokens and hashes) from the environment rather than hardcoding them in the source code.

social-media-intelligence