The Agent Skills Directory

[DATA_EXFILTRATION]: The skill accesses the sensitive configuration file ~/.zshenv using grep to extract the TAVILY_API_KEY.- [COMMAND_EXECUTION]: The skill instructs the agent to modify the user's shell profile (~/.zshenv) by appending sensitive API key exports. This serves as a persistence mechanism and risks exposing secrets in command history or shell configuration files.- [COMMAND_EXECUTION]: Arbitrary Python code is executed via python -c and within a virtual environment created in /tmp.- [EXTERNAL_DOWNLOADS]: Fetches and installs dependencies from official package registries (including gpt-researcher and langchain modules) to set up the research environment.- [PROMPT_INJECTION]: Indirect prompt injection surface identified as the skill processes untrusted content from the web and local documents while possessing file-write and network capabilities.
Ingestion points: SKILL.md describes ingesting web search results and local files (PDF, DOCX, etc.) from paths specified in DOC_PATH.
Boundary markers: Absent; no clear delimiters are used to separate untrusted data from instructions.
Capability inventory: The skill can execute shell commands, write files (e.g., background.md), and initiate network requests to various LLM and search providers.
Sanitization: No evidence of input validation or content sanitization before processing external data.

gpt-researcher