The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to install its core dependencies by piping a remote script to the shell (curl -fsSL https://parallel.ai/install.sh | bash). This is a high-risk pattern that executes unverified code from an external server.
[EXTERNAL_DOWNLOADS]: The skill downloads installation scripts and tools from https://parallel.ai and fetches the parallel-web-tools package via pipx from PyPI.
[COMMAND_EXECUTION]: The skill performs numerous shell operations using parallel-cli where user-provided variables (like $OBJECTIVE, $URL, and $QUESTION) are interpolated into command strings. This presents a potential command injection risk if the variables are not strictly sanitized.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and summarize live web content, which could contain adversarial instructions aimed at hijacking the agent's session.
Ingestion points: Live web data retrieved through parallel-cli extract and parallel-cli search commands described in SKILL.md.
Boundary markers: Absent. The instructions do not provide the agent with delimiters or specific warnings to ignore commands embedded within the fetched text.
Capability inventory: The skill has the ability to write files to /tmp/, initiate network requests (via monitoring and webhooks), and install packages.
Sanitization: Absent. There is no mention of filtering, escaping, or validating the content retrieved from external URLs before it is processed by the LLM.

parallel-ai-search