The Agent Skills Directory

[DATA_EXFILTRATION]: The skill accesses highly sensitive local data sources to perform its primary function. Specifically, scripts/preprocess.py is used to extract data from the macOS iMessage database (~/Library/Messages/chat.db) and WeChat SQLite databases. While the instructions state that all data is processed locally, accessing private messaging history constitutes a significant data exposure risk.
[DYNAMIC_EXECUTION]: The skill is designed to generate new AI agent skills. In Phase 6, it maps extracted persona dimensions to a persona.json file and uses npx openpersona create to generate a complete OpenPersona skill pack, which includes its own SKILL.md instructions and JavaScript scripts (scripts/state-sync.js). These generated skills are then intended to be installed and executed by the agent.
[INDIRECT_PROMPT_INJECTION]: The distillation process involves ingesting untrusted external data from various sources, including chat exports (WhatsApp, Telegram, Slack), social media archives (Twitter/X, Instagram), and web search results. The skill instructions do not specify robust boundary markers or sanitization to prevent malicious instructions hidden within these inputs from influencing the agent's behavior during the persona extraction and distillation phases.
[COMMAND_EXECUTION]: The skill frequently utilizes the Bash tool to perform administrative and data-processing tasks. This includes running Python scripts for database extraction (preprocess.py), managing persona versions (version_manager.py), and interacting with the OpenPersona CLI (npx openpersona).
[EXTERNAL_DOWNLOADS]: The skill references and uses the openpersona package via npx and integrates with other vendor-owned skills such as skills/open-persona and skills/persona-knowledge. These dependencies are part of the intended ecosystem provided by the vendor.

anyone-skill