The Agent Skills Directory

[DATA_EXFILTRATION]: The skill is designed to search through and extract information from highly sensitive personal sources, including Telegram chat history (via the tg.py script), email communications (via the himalaya CLI tool), and local CRM/contact records in the vault/ directory. This data is aggregated into a dossier, creating a risk of exposing private and confidential communications.
[INDIRECT_PROMPT_INJECTION]: The skill processes large volumes of untrusted data from external sources, such as YouTube transcripts, social media profiles, and web articles. This information is used for recursive self-evaluation and psychoprofiling without the implementation of explicit boundary markers or sanitization, which could allow malicious instructions embedded in scraped content to influence the agent's behavior.
[COMMAND_EXECUTION]: The skill includes a universal Apify actor runner (run-actor.sh and run_actor.js) that allows the agent to execute any remote actor by ID. If the selection of these actors is influenced by external or untrusted inputs, it could lead to the execution of malicious remote logic or unauthorized data exfiltration.
[DATA_EXFILTRATION]: Research output is saved to the local /tmp/ directory, and the actor runner facilitates writing data to arbitrary file paths via the --output command-line argument, which could potentially be used to overwrite local files or expose sensitive data stored on the filesystem.

osint