The Agent Skills Directory

[SAFE]: The skill is designed for research purposes and exhibits safe handling of credentials and external data. It uses the Python standard library to communicate with official API endpoints for OpenAI, xAI, and Reddit. It implements secure configuration management by storing keys in a dedicated directory with restrictive file permissions.- [PROMPT_INJECTION]: The skill processes external data from Reddit and X, which inherently presents a surface for indirect prompt injection. However, this risk is well-managed through robust instructions that separate data gathering from analysis. * Ingestion points: Untrusted content enters the context via OpenAI and xAI API responses containing community discussions. * Boundary markers: The SKILL.md provides explicit grounding instructions for the Judge Agent to rely on extracted research patterns rather than general knowledge. * Capability inventory: The agent utilizes the Bash tool to execute the research script, Write and Read for managing cached reports, and WebSearch for supplementary data. * Sanitization: External content is parsed, normalized, and scored by Python scripts before being summarized for the agent, providing a layer of processing between untrusted data and the agent's core instructions.

last30days